Python: convert Pydantic model class response_format to JSON schema in OllamaChatClient

[anneheartrecord]

Motivation and Context

Closes #6781.

OllamaChatClient forwarded response_format straight to Ollama's format param without converting a Pydantic model class to its JSON schema. Ollama's format only accepts '', 'json', or a JSON-schema dict, so passing a model class — the form OpenAIChatClient/FoundryChatClient accept, and the form create_harness_agent's plan mode uses — raised a ValidationError while building the request, before any call reached Ollama. This made structured output via a model class unusable on Ollama and broke harness plan mode on every Ollama model.

Description

In _prepare_options, when mapping response_format → format, convert a Pydantic model class to its JSON schema (model_json_schema()), keeping the existing '' / 'json' / dict pass-through. The original response_format class is left untouched in options, so typed parsing of the response (ChatResponse.value → model instance) still works — matching the OpenAI/Foundry clients and Ollama's documented structured-output usage.

Added test_cmc_response_format_pydantic_model: asserts the outgoing format kwarg is the JSON schema dict (not the class) and that the response parses back into the model instance.

Contribution Checklist

• The code builds clean without any errors or warnings
• I have added tests for my changes (added test_cmc_response_format_pydantic_model; full ollama suite green)
• ruff format/check, pyright, and mypy pass on the changed files

[Copilot]

Pull request overview

Fixes structured-output support for the Python OllamaChatClient when callers provide options["response_format"] as a Pydantic model class (the same form accepted by other providers and used by harness plan mode). The client now converts the model class to a JSON Schema dict for Ollama’s format parameter while preserving the original class for typed response parsing.

Changes:

• Convert Pydantic model classes passed via response_format into model_json_schema() when translating to Ollama’s format request parameter.
• Update option documentation to explicitly describe support for Pydantic model classes.
• Add a unit test asserting the outgoing format is a JSON Schema dict and the parsed response is still an instance of the model.

Reviewed changes

Copilot reviewed 2 out of 2 changed files in this pull request and generated no comments.

File	Description
python/packages/ollama/agent_framework_ollama/_chat_client.py	Converts response_format model classes to JSON Schema for Ollama request format, preserving the class for typed parsing.
python/packages/ollama/tests/test_ollama_chat_client.py	Adds coverage for model-class response_format conversion + typed response parsing.