Add llama.cpp support for local OpenAI-compatible LLM backends

[vmlinuzx]

Summary

This PR adds first-class support for local llama.cpp-style LLM endpoints in the web UI.

What changed

• adds llama.cpp as an LLM provider preset
• routes llama.cpp through the existing OpenAI-compatible chat path
• makes API keys optional for local/OpenAI-compatible endpoints
• updates the settings UI to expose the new provider and clarify that API keys can be optional for local servers
• strips <think>...</think> reasoning blocks from assistant text before rendering/storing responses
• adds unit coverage for the new provider path and reasoning tag cleanup

Why

The app already supports several hosted providers, but local inference backends were awkward to use because:

• there was no provider preset for llama.cpp
• chat startup assumed an API key was always required
• some local/Qwen-family models may emit reasoning markup that should not be shown directly in the UI

This change keeps the existing provider architecture intact while making local OpenAI-compatible backends much easier to use.

Verification

Ran:

• pnpm --filter @openroom/webuiapps test -- --run src/lib/__tests__/llmClient.test.ts

Result:

• 46 tests passed