feat(otel): Add OpenTelemetry GenAI instrumentation to Copilot Chat

[zhichli]

Summary

Adds opt-in OpenTelemetry instrumentation to Copilot Chat following the OTel GenAI semantic conventions. Emits traces, metrics, and events for LLM calls, tool executions, agent orchestration, and embeddings. Existing telemetry (ITelemetryService) is unchanged.

What's included

Phase 0 — Foundation

• IOTelService interface + OTelServiceImpl (Node) with DI registration
• NoopOtelService for disabled/test/web paths
• Config resolver with layered env precedence (COPILOT_OTEL_* > OTEL_* > VS Code settings > defaults)
• GenAI semantic convention constants (genAiAttributes.ts)
• Message formatters for OTel GenAI JSON schema
• Metric instruments (gen_ai.client.token.usage, gen_ai.client.operation.duration, copilot_chat.*)
• Event emitters (gen_ai.client.inference.operation.details, session/tool/agent events)
• File exporters (JSON-lines fallback for CI/offline)
• OTLP HTTP + gRPC + console exporter support

Phase 1 — Wiring into chat extension

• Inference spans (chat {model}) in chatMLFetcher.ts — model, tokens, TTFT, finish reasons
• Tool spans (execute_tool {name}) in toolsService.ts — tool name/type/id, args/results (opt-in)
• Agent spans (invoke_agent {participant}) in toolCallingLoop.ts — parent span for full hierarchy
• Embeddings spans (embeddings {model}) in remoteEmbeddingsComputer.ts
• Content capture — full messages, responses, system instructions, tool definitions (opt-in via COPILOT_OTEL_CAPTURE_CONTENT=true)
• Metrics recording at all instrumentation points
• Diagnostic exporter logs first successful export for easy verification

Activation

Off by default. Enable via env vars:

COPILOT_OTEL_ENABLED=true
OTEL_EXPORTER_OTLP_ENDPOINT=http://localhost:4318

Respects telemetry.telemetryLevel — globally disabled when telemetry is off.

Span hierarchy (Agent mode)

invoke_agent copilot                    [INTERNAL]
  ├── chat gpt-4o                       [CLIENT]
  ├── execute_tool readFile             [INTERNAL]
  ├── execute_tool runCommand           [INTERNAL]
  ├── chat gpt-4o                       [CLIENT]
  └── ...

Testing

• 63 unit tests across 6 test files covering config, formatters, events, metrics, file exporters, noop service
• Verified E2E with local Jaeger (OTLP HTTP on :4318)

Risk

• Bundle size: OTel deps added (~200KB budget)
• Zero overhead when disabled (noop providers)
• No changes to existing ITelemetryService code paths