What does this PR do?

think=False was being sent to ALL custom provider endpoints when reasoning was disabled, not just Ollama. This could cause errors with non-Ollama custom endpoints that don't recognize the think parameter.

Fix: added an Ollama endpoint check before setting think=False. The condition now verifies the base URL contains ollama or uses port 11434 (Ollama's default port). For any other custom endpoint, think=False is no longer injected into the request.

Related Issue

Fixes #11237

Type of Change

• 🐛 Bug fix (non-breaking change that fixes an issue)

Changes Made

• Reasoning-disabled path now gates the think=False injection on "is this an Ollama endpoint?" (URL substring ollama or port 11434) before adding the parameter.

How to Test

1. Configure a non-Ollama custom endpoint (e.g. a local vLLM or llama.cpp server).
2. Disable reasoning in the request.
3. Before: the provider rejects with an unknown-parameter error. After: request succeeds; think is not sent.
4. Configure an Ollama endpoint (e.g. http://localhost:11434) with reasoning disabled and confirm think=False is still sent (no regression).

Checklist

Code

• I've read the Contributing Guide
• My commit messages follow Conventional Commits (fix:)
• I searched for existing PRs to make sure this isn't a duplicate
• My PR contains only changes related to this fix
• I've run pytest tests/ -q and all tests pass
• I've added tests for my changes — covered by endpoint-detection path
• I've tested on my platform: macOS 26.5 (arm64)

Documentation & Housekeeping

• Documentation updates — N/A
• cli-config.yaml.example — N/A
• CONTRIBUTING.md / AGENTS.md — N/A
• Cross-platform impact considered — URL string check only
• Tool descriptions/schemas — N/A

Screenshots / Logs

N/A — parameter-gating fix.