Talkinghead next6, also vector DB embeddings

[Technologicat]

Here, have some more postproc filters.

• shift_distort: Scifi depiction of a bad digital video transport. Random blocks of lines glitchily shift left and/or right. Usage instructions in README.
• analog_distort: Fluttering, runaway hsync near the upper or lower edge of the image. I've seen this caused by an overly long YPbPr component video cable between some equipment.

I originally thought I wasn't going to touch the ST-extras backend for a while until I have the docs ready, but if I'll be experimenting with adding an embeddings provider for RAG next, might as well send these talkinghead postproc filters first.

I'll return to finishing up talkinghead once I'm done with the RAG interlude.

[Technologicat]

Meh, the vectorizer backend was so simple to implement that I'm dumping it into this PR.

• Add endpoint /api/embeddings/compute, which computes the vector embedding for one or more sentences. This is meant for Vector Storage.
  • Using this requires loading a new module, "embeddings".
  • The text embedding model is chosen the same way as previously, using --embedding-model.
  • There is no need to load the "chromadb" module.
  • This does not need the chromadb Python library at all. Only the sentence_transformers library is needed.

[Technologicat]

Oh, and the vectorizes accepts batches, too.

EDIT: The matching frontend changes for Vector Storage are in the PR https://github.com/SillyTavern/SillyTavern/pull/1737

[Technologicat]

The content of this backend PR is now final.

Any new features can go into a new PR.