Add OpenAI-compatible tool call interop helpers

[Dominic789654]

Summary

• add tool-use helpers to parse XML-style Qwen/Tinker tool calls from assistant content into structured ToolCall objects
• add a helper to linearize assistant.tool_calls and role=tool history into plain assistant text for providers that reject OpenAI tool history
• document the interop use case and add unit tests

Context

This came from integrating Tinker-hosted Qwen-style models with OpenAI-compatible agent/eval frameworks. Some routes emit model-native XML tool calls in message.content instead of filling message.tool_calls, while subsequent chat calls may reject standard OpenAI tool-history messages. These helpers keep that adapter logic small and reusable.

A standalone extraction of the same idea is here for reference: https://github.com/Dominic789654/tinker2openai-tool

Tests

• python -m pytest tinker_cookbook/tool_use/openai_compat_test.py
• python -m ruff check tinker_cookbook/tool_use/openai_compat.py tinker_cookbook/tool_use/openai_compat_test.py tinker_cookbook/tool_use/init.py
• pyright tinker_cookbook/tool_use/openai_compat.py tinker_cookbook/tool_use/openai_compat_test.py

Note: uv run appeared to hang while initializing the fresh local environment, so I ran the checks through the existing Python/pyright executables instead.