Add AI search with pgvector

[matthewwong525]

Is your feature request related to a problem? Please describe. Allows people to resurface notes that are similar & allow people to search notes with AI.

Describe the solution you'd like

• Basically, on save (frontend), a hash will be created pushed into the note text hash table & an embedding will be created if the hash does not exist.
• Embedding will be created probably with the open AI api
• when searching an input will be text & the notes will be searched through the function.
• When you enable the plugin, the plugin will backfill the hashes.
• [ ] remove dio dependency

two tables:

• embeddings table (hash (id) , embeddings), hash as key
• note text hash table (auto updated, 3 columns: hash, note_column, note_id) - note_id / note_column as key

two funcitons:

• one that takes a supabase_json object(s) & creates the embeddings / updates the hash table (uses AI)
• one that takes a string as input (+ user credentials) & returns a list of all note ids with similarity scores.

Describe alternatives you've considered

• use locally hosted text embedders on deno (this is a lot more work)
• make sure to update the delete-user to also delete the note text hash table.

Additional context text embedding comparison article

[matthewwong525]

Use gte-small for text embeddings with deno functions. See: https://huggingface.co/Supabase/gte-small