fix(azure): promote x-ms-served-model header into Response.model for Responses API

[xodn348]

Summary

• Adds _process_response overrides to both AzureOpenAI and AsyncAzureOpenAI that promote the x-ms-served-model response header into Response.model for Responses API calls.
• Mirrors the existing behaviour for Chat Completions (where Azure already rewrites model from the header) so callers can see the actually-deployed model name rather than the logical alias.
• Zero impact on non-/responses paths and on responses that lack the header.

Issue

Fixes #3271 — when using AzureOpenAI.responses.create(...) the returned Response.model reflects the requested model alias (e.g. "gpt-4o") rather than the precise deployment actually served (e.g. "gpt-4o-2024-11-20"). Azure surfaces the real model name in the x-ms-served-model HTTP response header; this PR reads that header and writes it back into Response.model.

Changes

src/openai/lib/azure.py

• Added Type to the typing imports.
• Imported ResponseT from .._types.
• Added AzureOpenAI._process_response (sync): calls super(), then if the URL starts with /responses and x-ms-served-model is non-empty, patches result.model (non-streaming) or wraps stream._iterator to patch event.response.model for every event that carries a response: Response field (streaming).
• Added AsyncAzureOpenAI._process_response (async): identical logic using await super() and an async generator.

tests/lib/test_azure.py

• Added import for openai.types.responses.response.Response.
• Added five focused unit tests covering: header promoted (sync), header absent (sync), non-/responses endpoint not affected, header promoted (async), header absent (async).

Local verification

=== LOCAL_TEST_PASSED ===
$ python -m pytest tests/lib/test_azure.py -v
...
collected 64 items

tests/lib/test_azure.py::test_implicit_deployment_path[client0] PASSED
tests/lib/test_azure.py::test_implicit_deployment_path[client1] PASSED
tests/lib/test_azure.py::test_client_copying[sync_client-copy] PASSED
tests/lib/test_azure.py::test_client_copying[sync_client-with_options] PASSED
tests/lib/test_azure.py::test_client_copying[async_client-copy] PASSED
tests/lib/test_azure.py::test_client_copying[async_client-with_options] PASSED
...
tests/lib/test_azure.py::test_azure_responses_x_ms_served_model_promoted PASSED
tests/lib/test_azure.py::test_azure_responses_x_ms_served_model_absent PASSED
tests/lib/test_azure.py::test_azure_non_responses_endpoint_not_affected PASSED
tests/lib/test_azure.py::test_async_azure_responses_x_ms_served_model_promoted PASSED
tests/lib/test_azure.py::test_async_azure_responses_x_ms_served_model_absent PASSED
...
64 passed in 3.44s
=== LOCAL_TEST_PASSED ===

Risk

Low. The override only activates when:

1. The URL path starts with /responses (exact SDK-generated path for the Responses resource), AND
2. The x-ms-served-model header is present and non-empty.

All other paths and clients are untouched. Response uses a mutable Pydantic BaseModel (extra="allow"), so field assignment is safe. The streaming wrapper is a thin generator replacement on _iterator; it does not touch transport or parsing.