feat(jobs): unify execution supervision and history

[vishr]

Summary

• replace overloaded when with independent deployment_phase and operator_run policies
• make ob job run, ob job history, and ob job logs the single job-centric operator surface, with inputs sealed into plans
• gracefully stop timed-out containers with TERM, configurable shutdown_grace, then KILL, and retain forced-kill evidence
• show live phase, elapsed time, trigger, release, lock policy, attempt, and retry budget in status
• merge timer and operator records at read time and report last timer, last operator run, and last observed success
• make lock-contention diagnostics identify the scheduling rendezvous and point operators to live status

PR #190 is already in the base and provides the host-owned supervision used by operator-started scheduled jobs.

Breaking changes

• remove workloads.<job>.when
• add deployment_phase: none | pre_release | post_release
• add operator_run: allowed | disabled
• remove ob schedule run, ob schedule history, and ob schedule logs; use their ob job equivalents

Defaults preserve the useful intent without coupling the policies: phase none defaults to operator runs allowed, while release-phase jobs default to operator runs disabled.

Validation

• just check
• just lint
• focused regression coverage for graceful shutdown ordering, forced-kill evidence, live status, and unified history