The recent revision of the proposed MLBuffer spec no longer represents a low-level concept of device memory storage, as these buffers are now tied to specific shapes and data types. Converting between MLBuffer(s) of different sizes will likely require copying (vs reinterpret_cast). To avoid confusion with other 'buffer-as-bytes' types (e.g., GPUBuffer, ArrayBuffer), I propose renaming MLBuffer to MLTensor.

👍 = I vote yes to rename