Bug Description

The cron scheduler detects due jobs and advances their nextRunAtMs timestamp, but fails to actually execute/spawn the job sessions.

Environment

• OpenClaw version: 2026.2.3-1
• OS: macOS (Darwin 24.5.0 arm64)
• Node: v25.5.0

Steps to Reproduce

1. Create a cron job with schedule.kind: 'every' and short interval (e.g., 60000ms)
2. Wait for the job to become due
3. Observe that nextRunAtMs advances by the interval
4. But lastRunAtMs never updates - job never executes

Evidence

Test Job (1 min interval):
- Created at: 1770404809447
- First nextRunAtMs: 1770408960009
- After due time passed: 1770409020006 (+60 sec)
- lastRunAtMs: undefined (never ran!)

Position Monitor (2 min interval):
- nextRunAtMs before: 1770408960000
- nextRunAtMs after: 1770409080000 (+120 sec)
- lastRunAtMs: unchanged from hours ago

Expected Behavior

When a job's nextRunAtMs is reached, the scheduler should:

1. Execute the job (spawn session)
2. Update lastRunAtMs with execution time
3. Then advance nextRunAtMs

Actual Behavior

Scheduler skips step 1 and 2, only advances nextRunAtMs.

Workaround

Manual sessions_spawn works fine. Only the automatic cron scheduler loop is affected.

Additional Context

• cron status shows enabled: true with correct job count
• cron run <jobId> returns {ran: false, reason: 'not-due'} even when overdue
• cron wake doesn't fix it
• Gateway restart doesn't fix it
• Jobs with explicit model in payload are affected
• Both new jobs (Test Job) and existing jobs show same behavior