Cline v4.1.24 released: default model switched to Claude Sonnet 5.5

In v4.1.24, when no model is selected, the Cline and OpenRouter providers now use Claude Sonnet 5.5, which is also the first item in Cline's recommended list. The catalog adds Claude Haiku 5.5 plus providers like AI/ML API, LowRouter, and Prime Inference; Cline's free list gains Step 5 Preview; v0 offers only v0-1.5-md, and OpenCode Go switches to a new inference endpoint. Defaults for many providers (302.AI, GitHub Copilot, Google Vertex AI, Novita AI, etc.) change to Claude Haiku 5.5 or Muse Wonder 1.0, so users who haven't pinned a model will get different defaults. Fixes: disabling reasoning no longer returns 400 errors on GPT-6 Astra, GPT-6.1 Sol, Claude Fable 5, and Claude Opus 5.5, and reasoning-level-only Gemini models no longer error on thinking budget; MCP servers or hooks exiting early no longer crash the extension host; 404s for stale model ids no longer show as login issues but surface the real error; and output logs retain error details in production builds. Deleting slash commands from the input preserves surrounding text spacing, screenshots and images no longer inflate token estimates (which previously triggered auto-compaction too early), auto-compaction falls back to the selected model's window when a catalog entry has none, symlinked rule files load, and skills in .claude/skills are discoverable.