Make embedding database providers pluggable and add pg-vector

[thesteve0]

Adding a new vector datastore is currently a bit brittle, duplicates code, and doesn't leverage LlamaIndex's built in abstractions.

To address please:

1. Add some tests for db connectivity and storage - this will help to insure that refactoring doesn't break existing functionality.
2. Abstract out the vector data store connection to use a LlamaIndex pattern
3. Add some more config options for the vector data store in my_config.py - examples are tuning parameters for HNSW indices
4. Have Milvus use the new architecture
5. Add pg-vector as vector datastore.

[thesteve0]

Again, please assign to me