feat(google-genai): emit streaming latency span attrs

[sayonfortify]

Adds FortifyRoot streaming-latency span attributes to Google GenAI streaming responses.

What changed:

• generate_content_stream sync and async paths now pass request start_time into the existing streaming delegate.
• The streaming safety wrapper records:
  • fortifyroot.llm.streaming.time_to_first_token_ms when the first streamed item arrives.
  • fortifyroot.llm.streaming.time_to_generate_ms when the stream finalizes.
• Empty streams intentionally leave both attributes unset so RDS nullable columns remain honest no-data.
• Existing stream safety/finalization behavior is preserved.

Tests:

• Added deterministic sync/async unit coverage for TTFT/STTG attrs.
• Added empty-stream coverage proving no fake zero latency attrs are emitted.

Validation run locally:

• python3 -m py_compile on changed source/test files.
• PYTHONPATH=../opentelemetry-instrumentation-fortifyroot python -m pytest tests/test_safety_unit.py -q → 10 passed.
• git diff --check clean.

No network/provider calls; no cassette changes.