Use RawRepresentationFactory in Azure.AI.Inference embedding generators

[jozkee]

          > would it be appropriate to go with RawRepresentationFactory here or does the EncodingFormat concern is enough reason to revert?

We could write the code to successfully handle EmbeddingEncodingFormat.Float (and .Base64 is already handled), which should be trivially handled via deserialization. We would want to throw if it's set to other values, as they don't make sense with an Embedding<float> output. We can subsequently add overloads of AsIEmbeddingGenerator that produce other embedding types (e.g. BinaryEmbedding) if that's deemed valuable, but by far the most common use today is with float.

Originally posted by @stephentoub in #6433 (comment)

[jozkee]

Closing, we don't need to support other encoding formats at this time: #6445