Restore automatic execution for Agent task schedules

[Y1fe1Zh0u]

Summary

• start the legacy AgentSchedule scheduler in worker processes alongside the AgentTrigger daemon
• allow scheduled execution for active Agent states: creating, running, and idle
• stop running the one-time AgentSchedule to AgentTrigger migration from restart.sh
• add regression coverage for worker startup, active and stopped Agent states, and disabled Runtime handling

Root cause

The Agent Tasks UI still saves schedules as AgentSchedule, but worker startup only launched the daemon that scans AgentTrigger. The legacy scheduler was never started, so newly-created Tasks schedules remained stored but had no consumer when they became due. A startup migration could not solve this for schedules created after the service had already started.

Impact

Tasks created from the Agent Tasks page are consumed automatically again. The existing trigger subsystem remains unchanged, and stopped or expired Agents are still excluded from execution.

Timezone interpretation is intentionally not changed in this focused fix and will be handled separately.

Validation

• pre-fix regression: 3 expected failures reproduced
• focused scheduler/runtime tests: 18 passed
• complete backend test suite: 1993 passed
• Ruff on the changed scheduler and regression tests
• bash -n restart.sh
• git diff --check

Not tested

• deployment and wall-clock execution on the 3010 environment
• end-user timezone semantics