Vekterdb

Added sync_index_to_db() and train_index() to allow people a bit more flexibility to use the GPU. This allows you to set up an index using FAISS so that you can put what is needed onto a GPU. Then train it and then add the vectors from the database table into the index.

This represents a breaking change with the initial release. Two big changes were made

1. Dropped fetch_records() in favor of select() - In the select method you can specify any valid SQLAlchemy Where clause to determine which records to get. fetch_records only did a column.in_([0, 1, 2, 122, 55]).
2. Switched the serialize_vector() from using zlib to using numcodecs.zstd.compress() - Just utilizing normal speed test and you can also now provide a compression level to use.

Updated the tutorial to reflect these changes (and obviously the tests).

Initial release