FR: Copy OpenAI cache tokens to retry attempts

[sayonfortify]

Summary

• copy OpenAI response-body prompt_tokens_details.cached_tokens onto the canonical retry-attempt span as gen_ai.usage.cache_read_input_tokens
• copy total_tokens onto retry-attempt spans as llm.usage.total_tokens
• extend the retry-attempt unit test to assert the canonical attempt span carries total/cache usage attributes

Why

FortifyRoot backend now stores cache-token columns from span attributes, but live validation showed OpenAI cache tokens only on the skipped parent openai.chat span. The backend canonical usage event is the fortifyroot.openai.attempt_1 span, so cache usage must be present on that retry-attempt span to survive export batching and backend dedup.

Validation

• python3 scripts/apply-fr-modification-notices.py --check
• PYTHONPATH=../opentelemetry-semantic-conventions-ai:../opentelemetry-instrumentation-fortifyroot:. pytest tests/test_retry_attempt_emission.py -q from packages/opentelemetry-instrumentation-openai → 37 passed
• python -m compileall -q packages/opentelemetry-instrumentation-openai/opentelemetry/instrumentation/openai/retry_handler.py packages/opentelemetry-instrumentation-openai/tests/test_retry_attempt_emission.py

Downstream

After merge, cut the next fr-v0.52.6.N tag and vendor it into ocelle-py via ocelle-py/scripts/vendor.sh.