fix(langchain): stop overriding Azure request models

[Jackson Weber (JacksonWeber)]

Summary

• remove the distro-specific Azure callback-payload workaround now addressed by fix(openai): expose Azure deployment name in tracing and invocation p… langchain-ai/langchainjs#10886
• use the same request-first model selection for Azure and non-Azure LangChain runs
• preserve gen_ai.response.model independently from gen_ai.request.model
• update focused regression coverage and the changelog

Why a distro change remains

The distro currently detects bare AzureChatOpenAI runs and always prefers the server-reported response model for gen_ai.request.model. Once LangChain reports the configured deployment correctly, that workaround would overwrite the trustworthy request model.

This PR removes the Azure special case:

const effectiveRequestModel = requestModel ?? responseModel;

The response model remains populated separately when available.

Upstream dependency

This cleanup relies on the LangChain release containing langchain-ai/langchainjs#10886. That release must be used before shipping this change; older LangChain versions that report the unrelated gpt-3.5-turbo default will no longer receive a distro-side workaround.

Validation

• focused LangChain utility suite: 88 passed
• ESLint and Prettier checks passed for changed TypeScript files
• package build passed

[Jackson Weber (JacksonWeber)]

Wait to mark for review until langchain-ai/langchainjs#10886 is merged/released.