[FEATURE] Supporting fp16, byte and binary Index with GPUs

[navneet1v]

Description

This issue tracks the effort for adding the support for fp16, byte and binary index with GPUs. In the first release we are only enabling the fp32 based support but since k-NN support different types of quantization techniques ideally, we should support all different quantizations with GPUs.

Quantization/Data Types to be supported

1. fp16: Supported using faiss encoders
2. byte : Supported using faiss encoders and also as a datatype.
3. binary : Separate index type.

I have created a GH issue on faiss to see how we can support these types with GPUs facebookresearch/faiss#4223