Expose Claude Code context-scaling on /api/status (#163)

[seikixtc]

Summary

Closes #163.

The issue asked for an API endpoint that statuslines (Claude Code, custom prompts) can poll to render oMLX-specific stats — context limit, usage, cache efficiency. oMLX already ships a GET /api/status endpoint that covers most of this, but it's missing the Claude Code context-scaling fields and isn't documented in the README, so users don't know it exists.

This PR closes the gap with a minimal change:

1. Surfaces claude_code_context_scaling_enabled and claude_code_target_context_size on /api/status, matching the flat naming already used by the admin /api/stats response.
2. Documents /api/status in the README under "API Compatibility", with a curl example and a representative response payload, so statusline integrations can discover it.

No new endpoint, no new auth path, no schema break.

Files changed

Area	File
Server	omlx/server.py (/api/status response)
Tests	tests/test_status_endpoint.py (two new test cases)
Docs	README.md (new "Status API" subsection)

Test plan

• pytest tests/test_status_endpoint.py -v — all existing cases plus two new ones:
  • test_claude_code_fields_default_when_no_settings — verifies safe defaults (False, 200000) when global_settings is None.
  • test_claude_code_fields_reflect_settings — verifies configured values flow through (True, 131072).
• Manual:

  omlx serve --model-dir ~/models &
  curl -s http://localhost:8000/api/status | jq .claude_code_context_scaling_enabled
  

Sample response

{
  "status": "ok",
  "version": "0.3.8",
  "uptime_seconds": 12834.2,
  "default_model": "Qwen3-Coder-Next-8bit",
  "loaded_models": ["Qwen3-Coder-Next-8bit"],
  "cache_efficiency": 87.5,
  "claude_code_context_scaling_enabled": true,
  "claude_code_target_context_size": 200000,
  "...": "..."
}

Notes for review

• I went with flat field names (claude_code_context_scaling_enabled) rather than a nested claude_code: {...} block to stay consistent with /api/stats. Happy to switch to nested if you'd rather have a cleaner schema for external consumers — easy to flip.
• The README example response shows cache_efficiency and the other already-exposed fields so a reader can see the full statusline payload in one place. If you'd prefer a minimal example that only highlights the new fields, let me know.
• No version bump — assuming this rides whatever release goes out next.