fix(cron): prevent migration repair from swallowing manual triggers (#78516)

[blut-agent]

Problem

When trigger_job() sets next_run_at to _hermes_now().isoformat() (essentially 'now'), the migration repair branch in _get_due_jobs_locked() can recompute next_run_at from the cron expression and skip the job — swallowing the dashboard 'Run now' button when there's a timezone offset mismatch.

This was reported in issue NousResearch#78516 where the dashboard 'Run now' button appeared to do nothing because the scheduler tick would detect a timezone offset mismatch and recompute the next run time, effectively skipping the manual trigger.

Fix

Add a proximity check (_MANUAL_TRIGGER_PROXIMITY_SECS = 60): if next_run_dt is within 60 seconds of now, treat it as a manual trigger from trigger_job() and let the job fire. Genuine stale runs (>60s ago) still get the migration repair treatment.

Changes

1. cron/jobs.py: Added _MANUAL_TRIGGER_PROXIMITY_SECS = 60 constant and a proximity check in the migration repair branch of _get_due_jobs_locked()
2. tests/cron/test_jobs.py: Added TestManualTriggerProximity class with two regression tests

Tests

• test_trigger_not_swallowed_when_within_proximity: Verifies that a manual trigger (next_run_at within 60s of now) is NOT swallowed by the migration repair
• test_trigger_swallowed_when_stale: Verifies that genuine stale runs (>60s ago with offset mismatch) still get the migration repair treatment

All 56 cron tests pass.