[Observability 9/10] Add OTEL distributed tracing to MCP server

[concaf]

Context

Add OpenTelemetry distributed tracing with spans across the full MCP request lifecycle: HTTP request → auth middleware → tool handler → downstream calls (Snowflake, embedding, K8s). This provides end-to-end visibility for debugging slow or failed requests.

Scope (to be detailed)

• Add TracerProvider initialization to pkg/telemetry/
• Create spans in tool handlers
• Propagate trace context to downstream HTTP calls (embedding client) and database calls (Snowflake)
• Connect with the request_id already generated in logger.NewContext

Use cases enabled

E4 (full end-to-end trace correlation)

Depends on

• [Observability 1/10] Add OTEL metrics foundation — telemetry package + main.go wiring #308 (telemetry foundation)
• Issue 8 (log trace correlation) — recommended to do first so logs and traces align