Add tensor subclass to handle ggml quantized tensor (#2504)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/2504

Add a tensor subclass to hold ggml quantized q4_0 tensor
* from_float(): quantize function from fp16 to q4_0
* to_float(): dequantize

bypass-github-export-checks

Reviewed By: mergennachin, digantdesai

Differential Revision: D55042867

fbshipit-source-id: 1c78aa1f4039616a672847035e1d0c60b492615a