feat(models): add OpenAI Responses API adapters

[FrigaZzz]

Feature Request: Add OpenAI Responses API model adapters

Related issues

• adk-python #3209

Summary

Add community model adapters for the OpenAI Responses API, including Azure OpenAI Responses API support, under google.adk_community.models.openai_responses.

Problem

google-adk-community does not currently provide a model adapter for the OpenAI Responses API. Users who want to use OpenAI or Azure OpenAI Responses API models with ADK need to implement their own request conversion, response parsing, streaming handling, tool-call mapping, and usage metadata handling.

This creates duplicated integration work for users who want to run ADK agents against OpenAI Responses API-compatible models, especially Azure OpenAI deployments exposed through the /openai/v1/responses endpoint.

Proposed Solution

Add two model adapters:

• OpenAIResponsesLlm for OpenAI Responses API models.
• AzureOpenAIResponsesLlm for Azure OpenAI deployments that expose the OpenAI-compatible Responses API.

The adapters should support:

• Conversion from ADK LlmRequest into OpenAI Responses API responses.create parameters.
• Conversion of ADK contents, system instructions, tools, tool responses, and generation config into Responses API-compatible request shapes.
• Parsing OpenAI Responses API responses into ADK LlmResponse objects.
• Text output, function calls, reasoning parts, finish reasons, interaction IDs, model versions, and usage metadata.
• Streaming Responses API events, including partial text and reasoning deltas plus final response aggregation.
• Azure OpenAI-compatible client construction using the /openai/v1/ base URL.
• Optional OpenAI response metadata on LlmResponse.custom_metadata.
• An optional openai dependency extra so users can install the OpenAI SDK only when needed.

Example Usage

from google.adk_community.models.openai_responses import OpenAIResponsesLlm

model = OpenAIResponsesLlm(model="gpt-5")

from google.adk_community.models.openai_responses import AzureOpenAIResponsesLlm

model = AzureOpenAIResponsesLlm(
    model="my-gpt-5-deployment",
    azure_endpoint="https://my-resource.openai.azure.com/",
)

Alternatives Considered

• Keep this in core ADK: this feature is better suited for google-adk-community because it is a third-party model integration and can evolve independently from the stable core ADK package.
• Require users to call the OpenAI SDK directly: this does not integrate with ADK runners, sessions, tool calls, streaming, or LlmResponse semantics.
• Provide only an OpenAI adapter: Azure OpenAI users need first-class support for deployment names, Azure endpoint configuration, and the /openai/v1/ base URL.

Testing Requirements

Unit tests should cover:

• Request shape conversion.
• System instructions and generation config mapping.
• Tool declaration and tool response mapping.
• Typed and mapping-based response parsing.
• Function calls and function call IDs.
• Reasoning metadata.
• Usage metadata.
• Streaming aggregation.
• Failed stream handling.
• Azure base URL construction.
• Optional disabling of response metadata in custom_metadata.

Manual E2E validation should be possible with a sample Azure OpenAI Responses agent under contributing/samples/models/hello_world_azure_openai_responses.

Acceptance Criteria

• Users can import OpenAIResponsesLlm and AzureOpenAIResponsesLlm from google.adk_community.models.openai_responses.
• Users can install required dependencies with pip install "google-adk-community[openai]".
• ADK agents can use the model adapters with the existing runner flow.
• Tool calling works through ADK function tools and OpenAI Responses API function calls.
• Streaming and non-streaming generation both return valid ADK LlmResponse objects.
• Azure OpenAI deployments use the expected /openai/v1/ base URL.
• Unit tests pass locally and in CI.
• Documentation and a manual E2E sample are included.

[FrigaZzz]

#149 closes this issue