feat: add AI-powered cluster naming with fallback to TF-IDF

[Harsh16gupta]

This PR adds cluster naming using Joplin's native AI chat API (joplin.ai.chat()).

Changes

• Added aiNamingService.ts to format prompts, parse JSON responses, and sanitize category names.
• Integrated AI naming into runPipeline.ts right after TF-IDF tag extraction (runs for both native embeddings and local ONNX worker paths).
• Extended src/types/joplinAi.d.ts type declarations to include ChatMessage, ChatResult, and chat().
• Added unit tests in test/pipeline/clustering/aiNamingService.test.ts.

Fallback Behavior

• If AI is disabled, unconfigured, times out (>45s), or returns an invalid response, the pipeline falls back to existing TF-IDF names.
• postProcess.ts and clusterNaming.ts remain untouched.

Testing

naming after using ai chat api

[Harsh16gupta]

Regarding Provider Support:

joplin.ai.chat() routes through whatever provider the user has configured in Joplin's AI settings:

• Local models via Ollama, LM Studio, or LocalAI
• OpenAI-compatible APIs like DeepSeek, OpenAI, or Groq
• Joplin Cloud AI and Anthropic (Claude)

The pr does not hardcode any provider or API key; it uses whichever model the user has active in Joplin Desktop.

I have tested it with deepseek, @HahaBill do you want me to test with other providers?