feat(operators): wire metrics field and add collect_metrics() to OperatorManager

[bootcrowns]

Summary

OperatorManifest has had a metrics: List[str] field since the initial commit but OperatorManager never read it, leaving the monitoring infrastructure described in the roadmap (Workstream 1 – Metrics Collection) unimplemented.

This PR wires the metrics field through the manager in three places:

Changes

activate() — passes manifest.metrics into the scheduler task metadata so workers/plugins can introspect which metrics the operator declares.

status() — includes metrics in the per-operator status dict alongside tools, schedule_type, etc., so callers get a complete view of the operator's declared monitoring intent.

collect_metrics(operator_id, *, since, until) [new method] — queries system.telemetry (a TelemetryAggregator instance) and returns only the summary-level stat fields explicitly declared in manifest.metrics. Behaviour:

• Returns {} gracefully when manifest.metrics is empty or telemetry is not configured.
• Unknown metric names are skipped with a DEBUG log (forward-compatible with future stats).
• Supports optional since/until Unix-timestamp filters, passed through to TelemetryAggregator.summary().
• All telemetry errors are caught and logged, never raised to callers.

Supported metric names

Any field on AggregatedStats from openjarvis.telemetry.aggregator:
total_calls, total_tokens, total_cost, total_latency, total_energy_joules, avg_throughput_tok_per_sec, avg_gpu_utilization_pct, avg_energy_per_output_token_joules, avg_throughput_per_watt, total_prefill_energy_joules, total_decode_energy_joules, avg_mean_itl_ms, avg_median_itl_ms, avg_p95_itl_ms.

Example operator TOML

id = "monitor-agent"
name = "Monitor Agent"
schedule_type = "interval"
schedule_value = "300"
metrics = ["total_calls", "avg_gpu_utilization_pct", "total_energy_joules"]

Testing

• No existing tests were changed or broken (pure additive surface).
• collect_metrics is safe to call when system.telemetry is None (returns {}).

Relates to Workstream 1 (Metrics Collection) in the development roadmap.

[robbym-dev]

Hi @bootcrowns ,

Thank you for your contribution! Would you be able to repair the lint errors? Once fixed, I can merge the PR. Thank you so much!

[bootcrowns]

Hi @robbym-dev, thanks for the review! I've wrapped the long lines in collect_metrics to satisfy the E501 lint check. Ready for merge now. I've also pushed the requested fixes for #202. Thanks!