feat(openai): emit streaming latency span attrs

[sayonfortify]

Adds FortifyRoot span attributes for OpenAI streaming latency so TTFT/STTG can be extracted into RDS while leaving existing Mimir histograms unchanged.

Attributes:

• fortifyroot.llm.streaming.time_to_first_token_ms
• fortifyroot.llm.streaming.time_to_generate_ms

Implementation:

• Sets TTFT/STTG as integer milliseconds in ChatStream, _build_from_streaming_response, and _abuild_from_streaming_response.
• First-token detection is independent of histogram availability, so tracing/RDS span attrs still work when metrics histograms are disabled.
• Empty streams set neither attr, preserving nullable RDS semantics.
• Existing histogram names, units, and .record() calls are preserved.

Tests:

• New deterministic unit test drives the v0 streaming generator with fake chunks/stub histograms.
• Covers: streaming sets attrs + histograms still fire; metrics-disabled path still sets attrs; empty stream sets no attrs.

Validation run locally:

• python3 -m py_compile packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/shared/chat_wrappers.py packages/opentelemetry-instrumentation-openai/tests/test_streaming_latency_span_attrs.py
• PYTHONPATH=packages/opentelemetry-instrumentation-openai:packages/opentelemetry-semantic-conventions-ai:packages/opentelemetry-instrumentation-fortifyroot /Users/sayondutta/opt/anaconda3/envs/py3.12/bin/python -m pytest packages/opentelemetry-instrumentation-openai/tests/test_streaming_latency_span_attrs.py -q → 3 passed.

No tag/PyPI release in this PR.