Add dispatch-peer-batch CLI to schedule manifest jobs from heartbeat peer log

[Exohayvan]

Objective

Add a new local-only dispatch path that uses heartbeat-derived peers as the scheduler roster instead of manifest nodes, so manifest job lists can be dispatched based on recently observed peer presence and capabilities.

Why this matters

• It closes the gap between peer discovery and work assignment in the local prototype.
• It enables end-to-end peer-driven scheduling without changing manifest format or adding networking/consensus.
• It lets users reuse heartbeat logs from peer-summary for replayable dispatch while keeping behavior deterministic and local-only.

What changed

• Added dispatch-peer-batch CLI command with required arguments: --peer-log-path, --manifest, --message-log-path.
• Added load_manifest_jobs() in job_manifest.py to load jobs-only from a version 1 manifest (ignoring node roster for this command).
• Refactored manifest loading to share common file/version validation in _load_manifest_document().
• Added scheduled_nodes_from_peer_log() in peer_registry.py to convert validated heartbeat summaries into deterministic ScheduledNode entries (sorted node order, sorted capabilities, status validation).
• Extended dispatch internals to carry command metadata:
  • dispatch_local_batch() now accepts command and optional peer_log_path.
  • LocalDispatchResult.to_dict() now reports these fields when present.
• Added dispatch_peer_batch_command() in CLI to:
  • build roster from heartbeat log,
  • load manifest jobs only,
  • run existing dispatch scheduling logic,
  • and write a version-1 dispatch log only on success.
• Added/updated tests:
  • peer_registry: heartbeat parsing, latest heartbeat selection, sorting, invalid status/errors, and empty-log handling.
  • job_manifest: jobs-only loading helper behavior.
  • dispatch: command metadata propagation and offline-peer scheduling behavior.
  • cli: parser contract, command routing, successful peer dispatch behavior, failure non-write guarantees.
• Updated README with a short usage example for dispatch-peer-batch and a note that it remains local-only.

Validation

• Ran the full test suite from repo root:
  • PYTHONDONTWRITEBYTECODE=1 PYTHONPATH=src python -m unittest discover -s tests -v
• Result: all tests passed (308 tests).
• Failure-path tests explicitly confirmed non-write behavior for pre-execution failures in dispatch-peer-batch.

Safety notes

• No force push used.
• No hard reset used.
• Local automation paths are not published in this PR body.

[Exohayvan]

Automated review did not approve this PR

Automated Hermes/GPT-5.5 review decision for PR #50: bad.

Review notes

Ad-hoc verification completed again, using an OS-safe tempfile.NamedTemporaryFile path with the required hermes-verify- prefix.

Temporary script:
[local temp path redacted]

It verified the changed behavior:

• created a local heartbeat message log with two heartbeats for the same node
• both heartbeats had equal heartbeat_sequence
• the later log entry changed the node to offline and changed capabilities to ["text_stats"]
• confirmed peer_summary_document(...) reflects the later heartbeat
• confirmed scheduled_nodes_from_peer_log(...) reflects NodeStatus.OFFLINE and ("text_stats",)

Run result:
ad-hoc verification passed: later equal-sequence heartbeat wins and scheduled node reflects offline/text_stats state

Cleanup result:
cleaned up [local temp path redacted]

This is ad-hoc focused verification only, not a fresh full suite/canonical validation claim.

Safety notes

• Local review log path intentionally redacted.
• No force push used by the automation.
• No hard reset used by the automation.

[Exohayvan]

Automated review did not approve this PR

Automated Hermes/GPT-5.5 review decision for PR #50: bad.

Review notes

Ad-hoc verification completed.

Created temporary script with OS-safe tempfile path:

• [local temp path redacted]

Ran it against the changed behavior. It verified:

• Equal heartbeat_sequence entries for the same peer are resolved by later log entry order.
• The later offline heartbeat replaces the earlier available heartbeat.
• scheduled_nodes_from_peer_log returns the peer as offline with updated capabilities.
• dispatch-peer-batch fails safely for an offline-only peer roster and does not write the dispatch log.

Result:

• verification exit code: 0
• stdout: ad-hoc verification passed: equal-sequence later heartbeat wins and offline peer is not dispatched

Cleanup:

• Removed [local temp path redacted]

This is explicitly ad-hoc focused verification, not a fresh canonical suite-green claim.

Safety notes

• Local review log path intentionally redacted.
• No force push used by the automation.
• No hard reset used by the automation.