fix(frontend): use embedding dimensions instead of model default dimensions for nearest neighbor search (Fix #49 )

[hydrosquall]

Motivation

• Enable use of nearest neighbor search when working with scopes that use non-default dimensions (for example, 768 instead of 1536 for openai-text-embedding-3-small
• Fix [bug] Nearest neighbor search errors if using non-default feature size for a text embedding model #49

Changes

• Pass through embedding dimensions when available from frontend to Explore page. ( datasets/$DATASET/explore/$SCOPE, where $DATASET and $SCOPE are variables
• My IDE removed trailing whitespace from a bunch of lines. You can review this diff ignoring those parts using this link ( https://github.com/enjalot/latent-scope/pull/50/files?diff=split&w=1 ), or I can look into putting the trailing whitespaces back.

Testing

For new users: here's the nearest neighbor search box

• Tested the nearest neighbor search box using a dataset of 1000 points and vector size 256 following the local development instructions
• Noticed an unrelated issue (search errors if your dataset has fewer than 150 points), but that seems like a separate fix (the frontend could add a UI to let the user pass in a hard coded nearest neighbor count, or default to a fraction of the total row count).

[enjalot]

thank you, including this in 0.4.0!