feat(clustering): improve auto-K selection algorithm and simplify settings

[Harsh16gupta]

This PR improves the Auto-K clustering selection logic and enables Auto-K when Joplin Native AI embeddings are active. It also cleans up unused setting options from Joplin preferences.

What changed:

• Enabled Auto-K (kmeans-auto, kmedoids-auto) for Native AI embeddings instead of forcing fixed K=6.
• Updated Auto-K search range for N >= 20 from sqrt(N) to floor(N / 3) (capped at 15) to prevent under-clustering on medium/large note collections.
• Added a 0.025 silhouette tolerance margin during K selection to prefer finer, more useful note categories over coarse clusters.
• Added seed decorrelation (seed + k) across K sweeps and a single-cluster fallback (K=1) for degenerate inputs.
• Removed distance metric and random seed settings from Joplin Options UI to simplify preferences (text embeddings default to Cosine similarity; low-D UMAP coordinates automatically use Euclidean distance).

Testing:
I have manually tested the whole pipeline, and the results are great!