feat: add semantic search tool for UI5 documentation

[dobrinyonkov]

Add search_ui5_docs tool that performs vector similarity search
against pre-computed embeddings of UI5 docs and code snippets by meaning rather than keywords.

• Local ONNX model (all-MiniLM-L6-v2) for query embedding
• Consumes pre-generated embeddings from resources/embeddings/
• Embeddings generated via separate tool: ui5-docs-rag
• Returns ranked results with similarity scores

Runtime behavior:

• First search downloads ONNX model (~86 MB) from HuggingFace
• Model cached at ~/.local/share/semantic-search/models/
• First search: 7s-3min (depends on connection)
• Subsequent searches: ~100-300ms

Alternative approaches considered:

• Download on server startup: delays server readiness
• Bundle model with package: adds ~86 MB to npm package, enables offline use

[dobrinyonkov]

Closing for now, will look into central search solutions.