Support new embeddings models and dimensions (3-small, 3-large)

This adds support for any text-embedding-* model, and necessarily updates the OpenApi service to support dimensions as a parameter since the default number of dimensions from some models is not supported by vector databases yet (such as Pinecone).

So each Vector Database plugin can say which dimensions it supports and allows the site builder to choose that when configuring embeddings.

The QueueWorker and SearchForm both had a direct implementation of embedding generation via the OpenAI PHP Client rather than OpenApi service wrapper from the base module here, so that needed rework as well to support dimensions.