feat(provider): auto-detect Ollama context limits

[felipemadero]

Fixes #10759

Summary

• Auto-detect Ollama servers by checking if GET / returns "Ollama is running"
• Query model context limits via POST /api/show to get num_ctx from Modelfile parameters
• Config limit.context takes priority if set by user
• Falls back to 4096 (Ollama's default) if not specified
• Fix compaction threshold for models without limit.output: reserve 10% of context instead of hardcoded 32000

This enables context percentage display in the status bar for Ollama models without manual configuration, and fixes compaction triggering immediately on small context models.

[github-actions]

Thanks for your contribution!

This PR doesn't have a linked issue. All PRs must reference an existing issue.

Please:

1. Open an issue describing the bug/feature (if one doesn't exist)
2. Add Fixes #<number> or Closes #<number> to this PR description

See CONTRIBUTING.md for details.

[github-actions]

The following comment was made by an LLM, it may be inaccurate:

Based on my search, I found the following potentially related PRs:

1. PR Adding the auto-detection of ollama local with a variable for baseURL #3726 - "Adding the auto-detection of ollama local with a variable for baseURL"

   • Adding the auto-detection of ollama local with a variable for baseURL #3726
   • Potentially related as it addresses Ollama auto-detection, though focused on baseURL variable rather than context limits

2. PR feat(opencode): add auto model detection for OpenAI-compatible providers #8359 - "feat(opencode): add auto model detection for OpenAI-compatible providers"

   • feat(opencode): add auto model detection for OpenAI-compatible providers #8359
   • Related to auto-detection of providers, may overlap with Ollama detection logic

3. PR fix(session): warn when context window may be too small for tool calling #7422 - "fix(session): warn when context window may be too small for tool calling"

   • fix(session): warn when context window may be too small for tool calling #7422
   • Related to context window handling, though focused on warnings rather than auto-detection

These are the closest matches, but none appear to be exact duplicates of PR #10758. You should review PR #3726 and #8359 for any overlap in Ollama auto-detection implementation logic.