fix: support current LM Studio model metadata

[themizzi]

Summary

• support current LM Studio native /api/v1/models records, including xhigh reasoning and newer fields
• validate LLM and non-LLM records separately
• preserve unloaded and loaded context discovery
• improve validation and fetch diagnostics
• harden fixture parsing and sanitization
• add regression coverage for current payloads and malformed records

Root cause

The release candidate schema validated every record as an LLM and restricted reasoning options to off, on, low, medium, and high. Current LM Studio responses include embedding records without LLM-only fields and reasoning levels such as xhigh, causing whole-response validation to fail and OpenCode to fall back to stale catalog models.

Verification

• npm run validate
• 32 tests passed
• live smoke test discovered the actual LM Studio LLMs and excluded embeddings

This supersedes the fork-local PR: themizzi#1