feat(core): Add high-precision performance telemetry (Latency, TPS, TTFT) to LLM events

[Yash3561]

Why are these changes needed?

Current LLMCallEvent and LLMStreamEndEvent capture token counts but lack the temporal telemetry required for production-grade agent monitoring. To optimize agentic workflows and enforce Service Level Agreements (SLAs), developers need to measure:

1. TTFT (Time To First Token): Critical for evaluating user experience in streaming agents.
2. TPS (Tokens Per Second): Essential for benchmarking throughput across different inference providers.
3. End-to-End Latency: Required to identify bottlenecks in complex multi-agent orchestration loops.

This PR implements high-precision timing using time.perf_counter() to provide these metrics without breaking backward compatibility.

Related issue number

Closes #5790

Checks

• I've included any doc changes needed for https://microsoft.github.io/autogen/. (No public documentation changes required for this internal telemetry upgrade).
• I've added tests (if relevant) corresponding to the changes introduced in this PR.
• I've made sure all auto checks have passed (Ran black and ruff linting).

Technical Details

• Event Refactor: Added latency_ms, tokens_per_second, and ttft_ms optional fields to LLMCallEvent and LLMStreamEndEvent in logging.py.
• Client Integration: Integrated timing logic into OpenAIChatCompletionClient and AzureAIChatCompletionClient.
• Streaming Logic: Captured ttft_ms by measuring the interval between request initiation and the first yielded chunk containing content or tool calls.
• Precision: Utilized time.perf_counter() to ensure monotonic, high-resolution measurements immune to system clock adjustments.
• Testing: Implemented regression tests in python/packages/autogen-core/tests/test_logging_events.py verifying both data presence and safe handling of optional fields.