Fix: log a failing widget once instead of on every request

[SandObserver]

Summary

• Add api/src/failure-log.js: tracks failures per key and logs the first failure, a changed reason, a reminder every 10 minutes with the failure count and start time, and one info line on recovery. Entries untouched for an hour are dropped.
• Route widget-data refresh failures and recoveries through it, keyed by widget id and endpoint.
• Add a quiet option to fail() and use it for widget-data, so a cached failure served to each tab is not logged again.

Testing
All green. Added unit tests for the tracker and an integration test that five failed requests log one line.

Changelog

Changed

• Log a failing widget once, with a reminder every 10 minutes and a line on recovery, instead of on every request.

Checklist

• Stays within the project constraints: one container, no runtime dependencies, no build step (see CONTRIBUTING.md).
• Tests added or updated for new logic, and the full suite, lint, and typechecks pass.
• Docs updated if behavior or configuration changed.