fix(web): parallelize firecrawl extract fanout

[LeonSGP43]

Summary

• parallelize Firecrawl multi-URL extract calls instead of scraping each URL serially
• keep per-URL timeouts, policy checks, and result ordering intact
• add a regression test proving two slow URLs no longer serialize

Closes #47599

Testing

• python3 -m pytest tests/plugins/web/test_web_search_provider_plugins.py -q
• uv run --frozen ruff check plugins/web/firecrawl/provider.py tests/plugins/web/test_web_search_provider_plugins.py
• git diff --check

[teknium1]

Thanks for the focused parallelization work. The current-main premise is confirmed: plugins/web/firecrawl/provider.py:458-496 still serially awaits each Firecrawl scrape.

Problems

• This branch predates current main's redirect-SSRF guard (2e12401ed; plugins/web/firecrawl/provider.py:529-547). That guard still uses results.append(...); continue, but this refactor removes results in favor of extract_one(). Preserve the guard by converting it to return {...} during salvage, or unsafe redirected URLs will hit an undefined results reference.
• The new timing test patches the client and website policy but not is_safe_url. The provider calls it at plugins/web/firecrawl/provider.py:530, and tools/url_safety.py:384-388 performs DNS resolution. Patch it to return True so the test remains hermetic.

Suggested changes

• Convert the current redirect-SSRF branch to the same per-task return shape as the other early exits.
• Stub is_safe_url in the regression test.

Automated hermes-sweeper review.