Change: skein-observability #122

Why

Charter §17 requires structured telemetry, correlation across operations, metrics, and
diagnostic CLI commands, and the system must be diagnosable without the dashboard.

Research (R10) found the Go skein repo already has internal/metrics (per-provider/
model/stage efficiency accounting, D1 fleet-efficiency-metrics; throughput) and
internal/traces, but no OpenTelemetry in go.mod. This change adds OpenTelemetry
where suitable, an explicit correlation-ID convention across the planes, the charter §17
metrics, and the diagnostic CLI — building on the existing metrics/traces rather than
replacing them.

What Changes

• Adopt OpenTelemetry where suitable (traces + metrics), building on the existing
  internal/metrics and internal/traces.
• Correlate every operation through objective_id, work_item_id, execution_id,
  agent_id, session_id, turn_id, model_request_id, resource_lease_id,
  trace_id.
• Expose the charter §17 metrics (throughput, efficiency, claim latency, lease usage,
  error rates) and the diagnostic commands skein status|projects|agents|work|fleet| trace|doctor, reconciled with the existing Go skein CLI.
• The AHP ahp-otlp: telemetry channel is a natural fit for host telemetry.

Impact

• Affected specs: skein-observability (new, cross-cutting), skein-control-plane
  (correlation IDs), ahp-host (OTel telemetry channel), mission-control (metrics
  views).
• Additive telemetry; disabling the exporter restores prior behavior.