feat(4-5/4-6): audit logging + zero-data-retention headers

Audit Logging (story 4-6):
- Nouveau modèle AuditLog (userId, action, resource, metadata, ip, createdAt)
- Migration 20260529143000_add_audit_log appliquée
- lib/audit-log.ts : logAuditEvent (fire-and-forget) + logAuditEventAsync + getClientIp
- auth.ts : LOG LOGIN / LOGOUT / USER_CREATED sur chaque event NextAuth
- /api/chat : log AI_REQUEST avec tokens + byok flag dans onFinish
- /api/agents/run-for-note : log AI_REQUEST avec featureKey + noteId

Zero-data-retention (story 4-5):
- OpenAI provider : header OpenAI-No-Training: 1
- Anthropic provider : header Anthropic-No-Train: 1
- DeepSeek provider : header X-No-Train: 1

sprint-status: 4-5 et 4-6 → done

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

memento-note/app/api/agents/run-for-note/route.ts

@@ -2,6 +2,7 @@ import { NextRequest, NextResponse } from 'next/server'
 import { auth } from '@/auth'
 import { prisma } from '@/lib/prisma'
 import { checkEntitlementOrThrow, QuotaExceededError, incrementUsageAsync } from '@/lib/entitlements'
+import { logAuditEvent, getClientIp } from '@/lib/audit-log'
 
 type GenerateType = 'slide-generator' | 'excalidraw-generator'
 
@@ -107,6 +108,14 @@ export async function POST(req: NextRequest) {
     .then(({ executeAgent }) => executeAgent(agent.id, userId))
     .catch(err => console.error('[run-for-note] Background agent error:', err))
 
+  logAuditEvent({
+    userId,
+    action: 'AI_REQUEST',
+    resource: featureKey,
+    metadata: { agentId: agent.id, noteId, featureKey },
+    ip: getClientIp(req),
+  })
+
   return NextResponse.json({ success: true, agentId: agent.id, status: 'running' })
 }