fix(translators): ameliorer le respect des formats de fichiers pour Excel, PowerPoint et PDF

- Excel: elargissement intelligent des colonnes pour le texte etendu et alignement RTL des cellules

- PowerPoint: retour a la ligne et auto-fit anti-debordement, colonnes de tableaux en RTL

- PDF: protection multi-colonnes pour eviter les chevauchements et formatage hebreu

- Tests: validation de non-regression pour tous les formats

translators/pdf_translator.py

@@ -169,7 +169,19 @@ def _shape_rtl_multiline(
         shaped_lines = []
         for l in lines:
             try:
-                shaped_lines.append(get_display(arabic_reshaper.reshape(l)))
+                # Arabic/Persian/Urdu script requires cursive contextual shaping (arabic_reshaper) + bidi display
+                # Hebrew script only requires bidi display (get_display)
+                has_arabic = any(
+                    "\u0600" <= ch <= "\u06FF"
+                    or "\u0750" <= ch <= "\u077F"
+                    or "\uFB50" <= ch <= "\uFDFF"
+                    or "\uFE70" <= ch <= "\uFEFF"
+                    for ch in l
+                )
+                if has_arabic:
+                    shaped_lines.append(get_display(arabic_reshaper.reshape(l)))
+                else:
+                    shaped_lines.append(get_display(l))
             except Exception:
                 shaped_lines.append(l)
         shaped_result.append("\n".join(shaped_lines))
@@ -1103,10 +1115,9 @@ class PDFTranslator:
             if not assigned:
                 columns.append([block])
 
-        # For each column, set next_block_y by y-order within that column.
-        # A block whose expanded bbox would touch the next column-mate's
-        # y0 will be capped to that y0 - 2 (small visual gap).
-        for column in columns:
+        # For each column, set next_block_y by y-order within that column,
+        # and set next_block_x to prevent expanding across adjacent columns.
+        for c_idx, column in enumerate(columns):
             column.sort(key=lambda b: b["bbox"][1])
             for i, block in enumerate(column):
                 if i + 1 < len(column):
@@ -1114,6 +1125,18 @@ class PDFTranslator:
                 else:
                     block["_next_block_y"] = page_bottom
 
+                # Multi-column protection: find adjacent column to the right
+                # with overlapping vertical bounds
+                next_col_x0 = page_rect.x1 - margin
+                b_y0, b_y1 = block["bbox"][1], block["bbox"][3]
+                for other_col in columns[c_idx + 1:]:
+                    for other_b in other_col:
+                        if other_b["bbox"][1] < b_y1 and other_b["bbox"][3] > b_y0:
+                            cand_x0 = other_b["bbox"][0] - 4.0
+                            if block["bbox"][2] < cand_x0 < next_col_x0:
+                                next_col_x0 = cand_x0
+                block["_next_block_x"] = next_col_x0
+
     def _write_translated_block(
         self,
         page,
@@ -1177,10 +1200,9 @@ class PDFTranslator:
         page_rect = page.rect
         margin = 18
 
-        # Track B3.6: try a wider bbox that respects the page margin.
-        # For headings, also allow horizontal expansion because long
-        # translated titles often don't fit in the original width.
-        max_x1 = page_rect.x1 - margin
+        # Track B3.6: try a wider bbox that respects the page margin and adjacent columns.
+        next_block_x = block.get("_next_block_x", page_rect.x1 - margin)
+        max_x1 = max(original_rect.x1, min(page_rect.x1 - margin, next_block_x))
         expanded_h = fitz.Rect(
             max(original_rect.x0, page_rect.x0 + margin),
             original_rect.y0,