Add MLTensor explainer (#754)

* add MLTensor explainer

* use correct usages in Fibonacci example + mention a tensor copy is achievable with identify

* address bberhar feedback: part 1

* address bbernhar feedback: part 2

* address bbernhar feedback: part 3

* make importExternalBuffer() async (among other changes)

* Use GPUExternalBuffer with STORAGE usage flag

* s/sourceData/inputData

* remove open question about cross-GPU-device interop

* address domenic feedback

* inline the formerly-nested MLTensorUsage dict into MLTensorDescriptor

* don't explicitly state that cross-GPU imports will be supported

* add WGSL code

* Typo GPUImportendTensor

---------

Co-authored-by: Dwayne Robinson <[email protected]>

Author: a-sully