Add support for buffer storage tensors (#3684)

Summary:
Pull Request resolved: https://github.com/pytorch/executorch/pull/3684

## Context

Add support for tensors that use buffer storage, in preparation for quantization support. For more context, the initial versions of quantized operators will target buffer based tensors. This is because the primary use-case is LLMs, which may contain tensors that may exceed the texture limits.

Reviewed By: yipjustin

Differential Revision: D57577019

fbshipit-source-id: 370184fa05e21b4db18a43bdfd2d13a6ee4bf49e