Record "recover-check" change and task logs when health check is failing

[benhoyt]

Also record a task for every failure until the threshold, and close
the change (set it to Done) when the health check goes successful
again. The error message and any details is included in the task log.

This is similar to the "recover-service" change in #117,
but for health checks.

Fixes #103.

[benhoyt]

Per discussion with @hpidcock (who had discussed with @niemeyer at the product sprint), we're leaning towards reserving changes and tasks only for user-initiated actions, and we're likely to add an "events" system and use events for this instead. Moving back to draft.

[benhoyt]

Per above comment, we're likely to use events for this instead. Closing for now (can always reopen later if necessary).