Wire the Agent365 example onto the OpenTelemetry SDK surface

[heyitsaamir]

Summary

This is the top layer of the Python Agent365 OpenTelemetry stack, building on the automatic baggage bridge in #516 and the token-provider surface in #530. It makes those SDK seams runnable in the Agent365 example and mirrors the app-owned integration demonstrated by microsoft/teams.ts#679.

Decisions

• Microsoft OpenTelemetry remains app-owned. microsoft-opentelemetry is declared only by examples/agent365; no SDK package imports it or acquires an exporter dependency. The lockfile change is comparatively large because Microsoft OTel 1.3.6 brings its supported Azure Monitor, OTLP, web-framework, HTTP, logging, and OpenAI instrumentation set transitively into the example environment.
• Async token acquisition is bridged to the synchronous exporter contract. The example refreshes an Agentic App Instance token through app.token_provider before opening each Agent365 operation scope, then serves it from a synchronous in-memory resolver. This keeps credential ownership on the existing SDK token surface without blocking inside the exporter callback.
• Correlation does not enable transcript capture. Sensitive-data export remains disabled, message bodies are not logged, and InvokeAgentScope is created without input or output content. The unused openai_agents instrumentation is explicitly disabled rather than adding an Agents SDK dependency.
• Short-lived proactive sends flush explicitly. The proactive entry point force-flushes after its scoped work so the batching exporter does not lose the final operation when the process exits.

Deferred

SDK token and baggage behavior stays in the lower stack layers (#530 and #516); this PR intentionally contains only the runnable example, its app-level dependency/configuration, documentation, and generated lockfile.

[corinagum]

this runs before _handle_message, so should failed oauth be logged so message handling can continue without telemetry?