Feat/add vector similarity search

[JohannesHool]

• Replace client-side fuzzy scoring with server-side search. get_datasets now takes a search_mode param: semantic (default) ranks the catalog via Huwise vector_similarity(), and lexical does a full-text search() match. Drops the old _tokenize/_match_score/_score_dataset logic and the difflib dependency.

• One API call instead of paging the whole catalog and ranking in-process; limit/offset are now honored server-side.

• Make the catalog configurable. DATA_PORTAL_DOMAIN is read from a committed .env next to main.py. Works with any Huwise portal; .env is force-included in the wheel build for easier testing with npx modelcontextprotocol/inspector.

• Normalize and escape query strings before building ODSQL

• README updated with the two search modes and catalog selection.