Instrument the runtime with Agent Framework OpenTelemetry traces, metrics, and logs

[KSemenenko]

Problem

The control plane needs built-in runtime observability so operators and developers can inspect agent, workflow, and tool behavior.

Scope

• Track OpenTelemetry traces, metrics, and logs aligned with Agent Framework patterns
• Cover agent runs, tool calls, workflows, and approval-related state
• Keep the default path local-first and product-usable

Out of scope

• Cloud exporter configuration beyond what the runtime must emit locally
• UI dashboards outside the local telemetry view issue

Implementation notes

• Follow Agent Framework observability guidance instead of inventing a custom tracing model
• Preserve separation between emitted telemetry and how it is visualized
• Keep sensitive-data handling compatible with later trust controls

Definition of Done

• The issue defines what the runtime must emit and why
• Later telemetry views and exports can build on a stable instrumentation plan

Verification

• Review the issue against the ADR and feature spec trust requirements

Dependencies

• Parent epic: [Epic] Add observability, evaluation, and trust controls #20