fix: bug with storing embedding info in metadata store

[sarahwooders]

Please describe the purpose of this pull request.
Previously the source.embedding_dim and source.embedding_model fields were always None because they were not being properly written during the load command. This resulted in errors during /attach which checks to make sure embedding models of the agent/source match.

How to test

1. Load: memgpt load directory --name memgpt_docs7 --input-dir docs --recursive
2. List sources memgpt list sources
3. Run /attach on memgpt run

Have you tested this PR?
Yes

[cpacker]

LGTM