Add aggregate wakeups for grouped remote jobs

[Copilot]

Manually stopped Bohrium jobs could become lost, while parallel jobs generated excessive per-job agent wakeups.

Changes

• Cancellation semantics

  • Normalize provider-reported stopped, terminated, or cancelled jobs to cancelled.
  • Record termination_origin=provider_observed without assuming who initiated termination.

• Durable job groups

  • Add session-scoped groups with an expected job count.
  • Support all-terminal, failure-ratio, and deadline wakeup policies.
  • Suppress individual lifecycle wakeups and emit one aggregate, restart-safe notification.

• Agent and harness integration

  • Expose group creation and group_id on Batch Job submission and attachment tools.
  • Include member outcomes and the triggering policy in the resumed agent turn.
  • Preserve session and explicit-control suppression behavior.

group = create_remote_job_group(
    name="parameter-sweep",
    expected_jobs=4,
    failure_ratio=0.5,
)

submit_bohr_batchjob(
    name="sweep-1",
    # ...
    group_id=group["group_id"],
)