Improve Ollama provider compatibility for v0.2.2

[HuRuilizhen]

Summary

This PR is a focused follow-up for v0.2.2 provider compatibility.

The main goal is to make the local Ollama path usable by default for reasoning-capable local models without changing the request behavior for the existing cloud providers.

Included

• disables reasoning by default for Ollama requests

  • adds reasoning_effort: "none" to Ollama chat-completions payloads
  • keeps the existing OpenAI-compatible provider path intact
  • does not change request payloads for OpenAI, Mistral, or Qwen

• adds focused regression coverage

  • verifies Ollama requests include reasoning_effort: "none"
  • verifies cloud-provider requests do not pick up the Ollama-specific field

Validation

Repository checks:

• ruff check .
• pyright
• pytest -q

Observed local-model check:

• reran examples/llm_extraction.py against a local Ollama service
• confirmed the example now completes successfully with qwen3.5:4b

Notes

• this PR intentionally keeps the compatibility fix local to the Ollama path
• it does not change the cloud-provider request contract for OpenAI, Mistral, or Qwen