fix(chat): parse structured output when reasoning returns chunked content

[Nelson-PROIA]

Problem

client.chat.parse() and parse_async() raise TypeError: Unexpected type for message.content: <class 'list'> whenever the response was produced with reasoning enabled, on any reasoning-capable model.

With reasoning enabled the API returns the assistant content as a chunked list rather than a plain JSON string:

"content": [
  {"type": "thinking", "thinking": [...]},
  {"type": "text", "text": "{...final JSON...}"}
]

convert_to_parsed_chat_completion_response in src/mistralai/extra/struct_chat.py only handled the str and None cases, so the chunked shape hit the else branch and raised.

Fix

When content is a list, concatenate the text of the TextChunk entries (ignoring ThinkChunk and any other chunk types) and validate that JSON against the caller's pydantic model, exactly like the string path. If the content contains no text chunks at all, parsed is set to None, mirroring the existing content is None behavior.

Tests

Two regression tests in src/mistralai/extra/tests/test_struct_chat.py:

• chunked content with thinking + text chunks parses into the pydantic model
• content with only thinking chunks yields parsed=None

Full local verification: unittest discover on src/mistralai/extra/tests (149 tests OK), pytest tests/ (358 passed, 46 skipped), plus mypy, pyright, and ruff on src/mistralai/extra/