Tuning embeddings

[gvanrossum-ms]

Try tuning embedding parameters depending on the embedding model.

For each embedding model we should determine optimal parameters for TextEmbeddingIndexSettings.

[gvanrossum-ms]

It should also be possible for users to provide their own tuning parameters for their own models.

But for well-known embedding models, I hope we can determine optimal tuning parameters (mostly min_score and max_hits) automatically, by taking an existing dataset (e.g. the Adrian Tchaikovsky podcast, for which we have a representative set of questions and expected answers), and measuring how successful it is with various values for the tunable parameters.