Make the embedding generation model we use configurable and use inference API

[dineshyv]

🚀 Describe the new functionality needed

Right now, embedding generation is hardcoded to a specific model, we want this to go through inference

💡 Why is this needed? What if we don't build it?

Needed for flexibility in using the correct model

Other thoughts

No response

[dineshyv]

cc: @raghotham , @ashwinb