FR: add request model to LLM metric attributes

[sayonfortify]

Description
This PR fixes a Metrics-tab consistency issue found during the Dashboard Provider Cost / GroupBy RCA validation.

After PR-A2 backend enrichment, Metrics Provider group-by correctly uses canonical fortifyroot.provider, but Metrics Model group-by can only canonicalize to the same model values as Events/Traces if metric datapoints include the request model. Several instrumentors only emitted gen_ai.response.model on metric datapoints. For routed/provider-backed calls, response model can be dated or provider-specific, while Events/Traces use request-model normalization.

Adds gen_ai.request.model to direct LLM metric emitters currently found in the fork: OpenAI, Anthropic, Google GenAI, Bedrock, and LangChain/LangGraph. LiteLLM/LlamaIndex are covered through underlying provider instrumentation when they do not emit native LLM metrics themselves.

What changed

• Adds gen_ai.request.model to LLM token/duration metric attributes for MVP metric emitters:
  • OpenAI
  • Anthropic
  • Google GenAI
  • Bedrock
  • LangChain / LangGraph OpenAI path
• Keeps existing gen_ai.response.model attributes unchanged.
• Does not touch SDK vendored files directly; this change belongs in the OpenLLMetry fork and will be vendored into the SDK after tagging.
• Does not add billing_provider; that remains PR-C scope.

Why
This lets backend metric enrichment use the same request-model semantics as LLM Events/Traces/Dashboard, so Metrics tab Provider/Model group-by does not drift into raw response-model labels like dated model IDs.

Validation

• git diff --check
• py_compile on all touched files
• Focused tests:
  • OpenAI metrics: 5 passed
  • Anthropic completion/message metrics: 4 passed
  • Google GenAI metric: 1 passed
  • Bedrock metric: 1 passed
  • LangGraph metric: 1 passed

Next
After merge, tag fork release fr-v0.52.6.14, then vendor that tag into fortifyroot-sdk-py.