Vendor OpenLLMetry fr-v0.52.6.21

[sayonfortify]

Summary

• vendors OpenLLMetry fork tag fr-v0.52.6.21 into Ocelle
• brings in FR: Copy OpenAI cache tokens to retry attempts openllmetry-py#32, which copies OpenAI retry-attempt usage fields onto the canonical attempt span
• specifically enables gen_ai.usage.cache_read_input_tokens to survive retry-attempt canonicalization so backend v0.0.80 can store OpenAI cache-read tokens in llm_usage_events

SOP / source provenance

• source repo: FortifyRoot/openllmetry-py
• source PR: FR: Copy OpenAI cache tokens to retry attempts openllmetry-py#32
• source tag: fr-v0.52.6.21
• source commit: 724081bb6f72
• vendored via ./scripts/vendor.sh ../fr-openllmetry-py . fr-v0.52.6.21

Validation

• OpenLLMetry fork PR checks passed across FR and non-FR suites for py3.10/3.11/3.12/3.13
• local fork validation before tag:
  • python3 scripts/apply-fr-modification-notices.py --check
  • PYTHONPATH=../opentelemetry-semantic-conventions-ai:../opentelemetry-instrumentation-fortifyroot:. pytest tests/test_retry_attempt_emission.py -q -> 37 passed
• local Ocelle validation:
  • PYTHONPATH=src pytest tests/openai/test_instrumentation.py -q -> 20 passed
  • PYTHONPATH=src pytest tests/test_ocelle_imports.py -q -> 3 passed
  • PYTHONPATH=src python -m compileall -q src/fortifyroot/_vendor/opentelemetry/instrumentation/openai/retry_handler.py
• live dev proof from the same SDK fix before vendoring: OpenAI cache probe wrote non-zero cache_read_tokens rows into RDS with backend v0.0.80 when canonical retry-attempt spans carried the cache attribute

Notes

• replacement for the closed direct-edit PR; this follows the required fork-first vendoring protocol
• no backend/proto/schema/frontend changes in this PR