fix(crawl): report the real timeout and stop crawling binary assets

[us]

Three fixes from production logs (2026-09-29 to 2026-10-13).

Timeout message reports the caller's budget

• When a render budget expired with nothing extractable, scrape_url_inner returned Timeout(fetch_result.elapsed_ms), the moment the partial render was taken. A request with timeout: 30000 answered after ~33s said Timeout after 2579ms; on the Firecrawl surface that text reaches the caller (SCRAPE_TIMEOUT). 292 timeouts in the window carried a 2.5 to 5s figure.
• Now Timeout(deadline.requested_ms()), as every other timeout path already does. The WARN line keeps elapsed_ms. Status codes are unchanged (v1 504, v2 408).

Crawl skips binary assets

• Link discovery takes every same-origin a[href], and gallery pages link straight to .jpg. Crawls fetched 638 image/video/archive links (250 MB in 34h), each read in full only to return an unsupported-content placeholder, and each used up a slot of the caller's limit.
• is_static_asset (last path segment's extension: images, media, archives, fonts, installers) skips them in crawl enqueue. pdf and office documents, which are parsed, are not in the list; neither are js/css, so slugs like /topics/node.js still crawl. Map still lists asset URLs but no longer fetches them for links.

Map discovery returns at its deadline

• The BFS awaited in-flight pages with no deadline of its own and only checked the clock after a page landed. A page whose recovery tier overran its budget held the loop open until the route's outer timeout fired and dropped every URL already found: 32 /v2/map responses were 504 with zero links at exactly 45s.
• The wait is now timeout_at(bfs_deadline_at, ...), so discovery returns what it found. New tests/discover_deadline.rs stalls a page body past the deadline: the old loop overshot by ~457ms, the new one by ~4ms.

No API shape change, no config change. Crawl results stop containing placeholder entries for asset links.