perf(audit): parallel Tesseract OCR + Pipeline-Wire-In für Slicing

ocr_slices_extract_cookies nutzt jetzt ThreadPoolExecutor (4 workers).
Tesseract released die GIL, daher echtes parallelisieren möglich.
Sequenziell 32 slices ≈ 60s, parallel ~15s.

Pipeline in agent_compliance_check_routes.py: Step C ruft jetzt
capture_cookie_evidence_slices + ocr_slices_extract_cookies. Source
'tesseract_ocr' wird zu existing Vendors gemergt; neue Vendors als
eigenständige Records.

Final VW-Scan-Resultat:
- Cookies: 60 (parse_flat) → 128 (mit Tesseract) = +113%
- Vendors: 18 unique
- Adobe Analytics: 9 → 33 Cookies (Tesseract fand +24)

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

backend-compliance/compliance/services/cookie_screenshot_ocr.py

@@ -413,38 +413,59 @@ async def capture_cookie_evidence_slices(
         return {"slices": [], "error": str(e)[:200]}
 
 
+def _ocr_one_slice(s: dict) -> tuple[dict, list[dict]]:
+    """Helper for parallel execution: tesseract + parse for one slice.
+    Returns (slice_metadata_summary, cookies)."""
+    import base64 as _b64
+    try:
+        png = _b64.b64decode(s.get("png_b64", ""))
+    except Exception:
+        return ({"idx": s.get("idx"), "ts": s.get("ts"),
+                 "top_y": s.get("top_y"), "bot_y": s.get("bot_y"),
+                 "cookies_found": 0}, [])
+    text = ocr_screenshot_via_tesseract(png)
+    chunk = parse_ocr_cookie_table(text)
+    return ({"idx": s.get("idx"), "ts": s.get("ts"),
+             "top_y": s.get("top_y"), "bot_y": s.get("bot_y"),
+             "cookies_found": len(chunk)},
+            chunk)
+
+
 def ocr_slices_extract_cookies(
-    slices: list[dict],
+    slices: list[dict], max_workers: int = 4,
 ) -> tuple[list[dict], dict]:
-    """Run Tesseract on each slice + parse + dedup by cookie name.
+    """Run Tesseract on each slice IN PARALLEL + parse + dedup by name.
+
+    Tesseract releases the GIL during its C-level OCR, so a
+    ThreadPoolExecutor with 4 workers yields ~4x speedup on multi-core
+    machines (M4 Pro has plenty). Sequential 32 slices = ~60s, parallel
+    ~15s.
 
     Returns (cookies, stats) where stats has:
-      per_slice: [{idx, cookies_found, ts}]
-      total_raw, total_unique
+      per_slice: [{idx, cookies_found, ts, top_y, bot_y}]
+      total_raw, total_unique, slices
     """
-    import base64 as _b64
+    from concurrent.futures import ThreadPoolExecutor
 
-    per_slice: list[dict] = []
+    if not slices:
+        return [], {"per_slice": [], "total_raw": 0,
+                    "total_unique": 0, "slices": 0}
+
+    # Keep slice order so the per-slice report is sequential.
+    with ThreadPoolExecutor(max_workers=max_workers) as ex:
+        results = list(ex.map(_ocr_one_slice, slices))
+
+    per_slice: list[dict] = [r[0] for r in results]
     all_cookies: list[dict] = []
     seen_names: set[str] = set()
-    for s in slices:
-        try:
-            png = _b64.b64decode(s.get("png_b64", ""))
-        except Exception:
-            continue
-        text = ocr_screenshot_via_tesseract(png)
-        chunk = parse_ocr_cookie_table(text)
-        per_slice.append({
-            "idx": s.get("idx"), "ts": s.get("ts"),
-            "top_y": s.get("top_y"), "bot_y": s.get("bot_y"),
-            "cookies_found": len(chunk),
-        })
+    for _, chunk in results:
         for c in chunk:
             nl = (c.get("name") or "").strip().lower()
             if not nl or nl in seen_names:
                 continue
             seen_names.add(nl)
             all_cookies.append(c)
+
     stats = {
         "per_slice": per_slice,
         "total_raw": sum(p["cookies_found"] for p in per_slice),
@@ -452,8 +473,8 @@ def ocr_slices_extract_cookies(
         "slices": len(slices),
     }
     logger.info(
-        "ocr_slices_extract_cookies: %d slices → %d raw → %d unique cookies",
-        stats["slices"], stats["total_raw"], stats["total_unique"],
+        "ocr_slices_extract_cookies (parallel=%d): %d slices → %d raw → %d unique",
+        max_workers, stats["slices"], stats["total_raw"], stats["total_unique"],
     )
     return all_cookies, stats