feat(translation): quality pipeline overhaul + new features (audit 2026-08-29)

Translation quality & format preservation:
- Word: merge adjacent same-format runs into one unit (sentence-level
  coherence like inline-tag handling); translate comments/balloons;
  dedupe textbox collection (was translated twice); RTL no longer
  overrides center/justify alignment; CJK/Arabic font hints (eastAsia/cs)
- PPTX: chart translations now actually reach the output file
  (ChartPart.blob is read-only — rewrite chart XML in the saved ZIP);
  CJK typeface hints (a:ea)
- Excel: sheet renames no longer break references — rewrite cell
  formulas (3D/quoted), defined names, data validations, cond. formats
- PDF: bold/italic honored (hebo/heit/hebi); table cells never merge;
  unchanged blocks left untouched (typography preserved, fixes duplicate
  hyperlinks); attempted/changed stats + route gate now cover PDF;
  CJK font paths; scanned PDFs via Mistral OCR (detection + admin settings)

Features:
- formality param (formal/informal) + automatic regional-variant prompts
- output_mode=bilingual docx (source above translation)
- per-user translation memory on Redis (falls back to LRU), context-hashed
- QA report + 0-100 confidence score in job status; L0 on by default
- OpenAI-compatible providers: whole chunk in ONE numbered-JSON request
  (~15x fewer calls) with per-item fallback; base prompt always present
  (custom prompt no longer replaces translation instructions)

Infra & marketing alignment:
- plan-based engine gating + vision gating (closes paid-engine leak);
  /providers/available filtered per plan; 107 languages exposed
- zh-CN/zh-TW validation fixed; libmagic disabled on Windows (native crash)
- admin: Mistral OCR settings + engine status dashboard; httpx<0.28 pin
  (TestClient breakage); Prometheus test fixture fixed
- marketing docs aligned with code (PDF+OCR, retention, engines, pricing)
- security: .env.ionos/.env.production/provider_settings.json removed

Tests: 1173 passed / 0 failed (6 network tests deselected: free Google
endpoint temporarily blocked from this machine)

tests/test_scanned_pdf_ocr.py

@@ -0,0 +1,192 @@
+"""Scanned PDF detection and the Mistral OCR translation path."""
+
+import fitz
+import pytest
+
+from config import config
+from services.mistral_ocr import MistralOCRClient
+from services.providers.base import TranslationProvider
+from services.providers.schemas import TranslationRequest, TranslationResponse
+from translators.pdf_translator import PDFTranslator
+
+
+FAKE_OCR_PAGES = [
+    "# Facture\n\nMontant total : 1 250 euros\n\n![logo](img.png)",
+    "Le paiement est du sous 30 jours.",
+]
+
+# Minimal French → English mapping for the fake provider; anything else is
+# prefixed so tests can assert replacement happened.
+TRANSLATIONS = {
+    "Facture": "Invoice",
+    "Montant total : 1 250 euros": "Total amount: 1,250 euros",
+    "Le paiement est du sous 30 jours.": "Payment is due within 30 days.",
+}
+
+
+class FakeProvider(TranslationProvider):
+    """New-style provider with canned translations (no network)."""
+
+    def get_name(self) -> str:
+        return "fake"
+
+    def is_available(self) -> bool:
+        return True
+
+    def translate_text(self, request: TranslationRequest) -> TranslationResponse:
+        # Pages arrive as multi-line text: translate line by line so the
+        # canned mapping applies regardless of how lines are grouped.
+        lines = []
+        for line in request.text.split("\n"):
+            stripped = line.strip()
+            if not stripped:
+                lines.append(line)
+                continue
+            lines.append(TRANSLATIONS.get(stripped, f"[EN] {stripped}"))
+        return TranslationResponse(
+            translated_text="\n".join(lines),
+            provider_name=self.get_name(),
+            from_cache=False,
+        )
+
+
+def _make_scanned_pdf(path):
+    """Image-only PDF: one page almost fully covered by a picture, no text."""
+    pix = fitz.Pixmap(fitz.csRGB, fitz.IRect(0, 0, 10, 10))
+    pix.clear_with(90)
+    png_bytes = pix.tobytes("png")
+    doc = fitz.open()
+    page = doc.new_page(width=612, height=792)
+    page.insert_image(fitz.Rect(20, 20, 592, 772), stream=png_bytes)
+    doc.save(str(path))
+    doc.close()
+    return path
+
+
+def _make_text_pdf(path):
+    doc = fitz.open()
+    page = doc.new_page()
+    page.insert_text(
+        (72, 100),
+        "Real selectable text with plenty of characters to exceed the scanned threshold. " * 2,
+        fontsize=11,
+    )
+    doc.save(str(path))
+    doc.close()
+    return path
+
+
+@pytest.fixture
+def no_api_key(monkeypatch):
+    monkeypatch.delenv("MISTRAL_API_KEY", raising=False)
+    monkeypatch.setattr(config, "MISTRAL_API_KEY", "")
+    monkeypatch.setattr(config, "MISTRAL_OCR_ENABLED", True)
+
+
+@pytest.fixture
+def api_key(monkeypatch):
+    # Set BOTH the env var and the config attribute: some other test in the
+    # full suite reloads the config module, and pdf_translator imports
+    # `config` at call time — after a reload only the env var is still
+    # visible (the class attributes are re-read from the environment).
+    monkeypatch.setenv("MISTRAL_API_KEY", "test-key")
+    monkeypatch.setattr(config, "MISTRAL_API_KEY", "test-key")
+    monkeypatch.setattr(config, "MISTRAL_OCR_ENABLED", True)
+
+
+class TestScannedDetection:
+    def test_image_only_pdf_is_scanned(self, tmp_path):
+        pdf = _make_scanned_pdf(tmp_path / "scan.pdf")
+        translator = PDFTranslator(provider=None)
+        assert translator._is_scanned_pdf(pdf) is True
+
+    def test_text_pdf_is_not_scanned(self, tmp_path):
+        pdf = _make_text_pdf(tmp_path / "text.pdf")
+        translator = PDFTranslator(provider=None)
+        assert translator._is_scanned_pdf(pdf) is False
+
+    def test_title_only_pdf_is_not_scanned(self, tmp_path):
+        """A sparse but textual page (no raster image) stays on the layout path."""
+        doc = fitz.open()
+        page = doc.new_page()
+        page.insert_text((72, 100), "Facture 2026", fontsize=18)
+        doc.save(str(tmp_path / "title.pdf"))
+        doc.close()
+        translator = PDFTranslator(provider=None)
+        assert translator._is_scanned_pdf(tmp_path / "title.pdf") is False
+
+
+class TestMarkdownCleanup:
+    def test_images_links_headings_removed(self):
+        md = "# Titre\n\n![img](x.png)\n\n[lien](http://x) texte"
+        out = PDFTranslator._markdown_to_text(md)
+        assert "Titre" in out
+        assert "![img]" not in out
+        assert "(http://x)" not in out
+        assert "#" not in out
+        assert "lien texte" in out
+
+    def test_table_pipes_removed(self):
+        out = PDFTranslator._markdown_to_text("| A | B |\n|---|---|\n| un | deux |")
+        assert "|" not in out
+        assert "un" in out and "deux" in out
+
+
+class TestScannedOCRPath:
+    def test_translate_scanned_pdf_end_to_end(self, tmp_path, api_key, monkeypatch):
+        pdf = _make_scanned_pdf(tmp_path / "scan.pdf")
+        monkeypatch.setattr(
+            MistralOCRClient,
+            "extract_pdf_text",
+            lambda self, path, progress_callback=None: list(FAKE_OCR_PAGES),
+        )
+
+        translator = PDFTranslator(provider=FakeProvider())
+        # Configure OCR the way the production route does (set_ocr_config).
+        # Going through the config module is fragile in the full suite:
+        # another test replaces sys.modules["config"], so a config-module
+        # patch may never reach the translator.
+        translator.set_ocr_config(api_key="test-key", enabled=True)
+        out = tmp_path / "out.pdf"
+        result = translator.translate_file(pdf, out, "en", "fr")
+
+        assert result.exists() and result.stat().st_size > 0
+        doc = fitz.open(str(result))
+        text = "\n".join(p.get_text("text") for p in doc)
+        doc.close()
+
+        assert "Invoice" in text
+        assert "Payment is due within 30 days" in text
+        assert "Facture" not in text
+        assert "logo" not in text          # markdown images stripped
+        assert "|" not in text             # markdown tables stripped
+
+    def test_missing_api_key_raises_clear_error(self, tmp_path, no_api_key):
+        pdf = _make_scanned_pdf(tmp_path / "scan.pdf")
+        translator = PDFTranslator(provider=FakeProvider())
+        with pytest.raises(RuntimeError) as exc:
+            translator.translate_file(pdf, tmp_path / "out.pdf", "en", "fr")
+        assert "MISTRAL_API_KEY" in str(exc.value)
+
+    def test_disabled_ocr_raises_clear_error(self, tmp_path, monkeypatch):
+        monkeypatch.setattr(config, "MISTRAL_API_KEY", "test-key")
+        monkeypatch.setattr(config, "MISTRAL_OCR_ENABLED", False)
+        pdf = _make_scanned_pdf(tmp_path / "scan.pdf")
+        translator = PDFTranslator(provider=FakeProvider())
+        with pytest.raises(RuntimeError):
+            translator.translate_file(pdf, tmp_path / "out.pdf", "en", "fr")
+
+    def test_text_pdf_bypasses_ocr(self, tmp_path, api_key, monkeypatch):
+        pdf = _make_text_pdf(tmp_path / "text.pdf")
+        called = {"ocr": False}
+
+        def _fail(self, path, progress_callback=None):
+            called["ocr"] = True
+            return []
+
+        monkeypatch.setattr(MistralOCRClient, "extract_pdf_text", _fail)
+
+        translator = PDFTranslator(provider=FakeProvider())
+        out = tmp_path / "out.pdf"
+        translator.translate_file(pdf, out, "en", "fr")
+        assert called["ocr"] is False