experimental proposal: add full text search as an alternative to semantic search

[ghost]

Right now, search for document chunks is performed in a "semantic" way using vector similarity. An alternative approach might be to use full text search (full text is already built into Postgres).

Changes required:

• rag/load.py: store the ts_vector along the text and embedding vectors

• rag/librag.py, function search() perform full text search (with stemming and stop words)

One could make full text search optional and if enabled the ranking is done by adding in a weighted way full text search rank and vector similarit search rank.

-- Chris Mair