[FEATURE] Support ray serve engine

[stikkireddy]

Ray Serve is a phenomenal serving engine that abstracts serving and some throughput optimization features like batching, async execution, pipelining, etc. Supports torch and other popular frameworks. This can be used for the following models:

1. custom embedding models with post processors
2. standard embedding models
3. encoder decoder models like whisper
4. diffusion models
5. multi model serving

[stikkireddy]

some common embedding models:

• Multimodal text-image embedding: clip-ViT-B-32
• Hebrew embeddings: dicta-il/dictabert-joint
• French embeddings: almanach/camembert-base