[georgeatparallel]

Problem

Search and Extract could lose usable results when another URL was excluded. The auth check also matched harmless author query params. Slow live fetches need to leave time for partial results.

Solution

• Stop matching auth inside author query params.
• Skip blocked URLs one at a time, keep usable results and errors, and set truncated when anything gets dropped.
• Give Extract a 20-second live-fetch timeout inside the existing 30-second client timeout. That leaves room for partial results, but it doesn't guarantee every request finishes.

Both fixes apply whenever Parallel is enabled. This targets #37's branch.

Testing

I ran three regression tests on the old code and the fix. They cover author URLs, mixed allowed and blocked results, and partial extraction. All three failed before and pass now. I didn't rerun the live Parallel/Browserbase checks.

Monitoring

I haven't verified runtime monitoring for omitted sources or Extract timeouts.