fix(scrape): make vision OCR a per-bundle allowlist, not a corpus sweep (#6)

Co-authored-by: claude <[email protected]>

scrape/runner.py

@@ -212,7 +212,10 @@ def scrape_toc_bundle(s: requests.Session, bundle: dict, force: bool, concurrenc
         page_html = fetch_toc_page(s, doc_id, entry.page_id)
         if not page_html:
             return False
-        body_md = html_to_md(page_html, s)
+        # Only pass the session (which enables image OCR) for bundles that
+        # opt in — image-only data lives in a few docs; the rest are UI
+        # screenshots that OCR would just turn into noise.
+        body_md = html_to_md(page_html, s if bundle.get("ocr") else None)
         sidecar = {
             "bundle_id": slug,
             "page_id": entry.page_id,
@@ -243,7 +246,7 @@ def scrape_single_bundle(s: requests.Session, bundle: dict, force: bool) -> int:
     if not html:
         print(f"  ! {slug}: empty body", file=sys.stderr)
         return 0
-    body_md = html_to_md(html, s)
+    body_md = html_to_md(html, s if bundle.get("ocr") else None)
     sidecar = {
         "bundle_id": slug,
         "page_id": doc_id,