docs(examples): add chat reasoning_effort examples

[andreaonofrei01]

Summary

Adds four runnable examples in examples/mistral/chat/ demonstrating reasoning_effort on mistral-medium-3-5. The SDK already supports the parameter and the ThinkChunk / TextChunk types, but there were no examples showing how to use them.

Files

• reasoning_response_shape.py : first stop. Calls the API once with reasoning_effort="high" and once with "none", then dumps the raw message.content so the reader can see the ThinkChunk / TextChunk JSON before consuming it.
• reasoning.py : single-turn call. Iterates message.content and branches on isinstance(chunk, ThinkChunk | TextChunk). Notes that reasoning_effort="none" returns a plain str.
• reasoning_with_streaming.py: the streaming version. Handles the three shapes that arrive on delta.content: ThinkChunk lists during the thinking phase, a transition list containing both a closing ThinkChunk and the first TextChunk, and plain string fragments after thinking ends.
• reasoning_multi_turn.py : 3-turn math chain (17 × 23 → × 3 → −100) run with two replay strategies (keep ThinkChunks vs drop them) and prints per-turn token usage so the cost difference is visible.

Notes

• All examples set httpx.Timeout(300.0) because reasoning runs routinely exceed the default 60s read timeout.
• Model is pinned to mistral-medium-3-5 (the model docs explicitly list as supporting reasoning_effort).

Test plan

• python examples/mistral/chat/reasoning_response_shape.py : verified output matches the documented ThinkChunk / TextChunk schema; effort="none" returns str.
• python examples/mistral/chat/reasoning.py : verified the John puzzle returns John is 22 years old.
• python examples/mistral/chat/reasoning_with_streaming.py : verified the trains problem returns 11:16:40 am and the thinking → final-answer transition renders cleanly.
• python examples/mistral/chat/reasoning_multi_turn.py : verified both strategies arrive at 1073; observed totals keep≈1656 vs drop≈637 tokens (varies with temperature=0.7).