Skip cron jobs with invalid schedules instead of 500-ing every cron mutation

[ramonski]

Context

The cron editor's pencil kept reverting a job's timeout to its template default (900), and edits appeared not to save. Root cause is a backend 500, not the UI.

hn-ai-daily had an out-of-range schedule 06 45 * * * (hour 45; max is 23). sync_jobs re-adds every enabled job on every cron mutation, and _add_job_to_scheduler (kaisho/cron/scheduler.py) only guarded the field-count check (_cron_kwargs), not CronTrigger's value validation. So that one malformed job made every cron enable/disable/update throw a ValueError → 500.

The mutation writes the file before sync_jobs, so the value did persist to disk — but the request 500'd, the frontend's React Query mutation errored, no refetch happened, and the edit dialog kept showing the stale cached value. That is the "timeout reverts to 900" symptom.

Why this matters

One bad job silently bricks all cron mutations, including the save that would fix the bad job — a deadlock the user can only escape by hand-editing jobs.yaml.

Fix

Validate the CronTrigger inside the guard in _add_job_to_scheduler, so a malformed schedule (wrong field count or out-of-range value) is logged and skipped instead of aborting the whole sync. The remaining jobs schedule normally and mutations return 200.

Adds tests/test_cron_scheduler.py covering: valid job registers, out-of-range hour is skipped without raising, wrong field count is skipped, and a bad job in a set does not block the good ones.

Out of scope

• Save-time schedule validation (a 400 with a clear message on POST/PATCH) so invalid schedules can't be stored in the first place — worth a follow-up.
• Surfacing skipped/invalid jobs in the UI (today they're only logged).