v316: Fix refresh after Firecrawl research cancellation

[nicu-chiciuc]

Admin research refresh failed when Firecrawl still reported a cancelled job as processing, then rejected cancellation with HTTP 409, Agent is already cancelled. Accept that exact response from the cancellation endpoint so refresh can start a new research job under the existing six-minute limit. Other conflicts and provider failures still block refresh, and completed results remain recoverable.

Return provider SDK failures as ConvexError with the method, path, message, HTTP status, and error code. The site page displays that detail instead of the generic production server error. Update the required hackathon log.

Validation:

• Reproduced the stale processing status and the exact HTTP 409 response against the previously cancelled Firecrawl job.
• Regression coverage verifies a fresh job can complete after that response, historical task research stays unchanged, unrelated conflicts still block refresh, and the UI displays provider details.
• pnpm run check passed: 1,668 tests passed, 11 skipped. No production deployment or new production research run.

---

^{Need help on this PR? Tag @codesmith-bot with what you need. Autofix is disabled.}