[RFC][TIR] Layout transformations on buffer access (#39)

* [RFC][TIR] Separate physical and logical layout of buffers

* Updated with link to the PR

* Fixed typo in example

* Added link to RFC#0040.

* Updated version with a function to define the physical layout.

* Updated with reference to iter_affine_map.h

* Updated following TQ's suggestions.

* Updated with link to BufferPointer RFC

* Updated following some comments from @vinx13

* Added description of the BufferTransform node.

* Added details on te.AXIS_SEPARATOR

* Updated buffer transformations to be in the PrimFunc::attrs

* Added author list, Eric/Wuwei

* Clarifying updates following comments from @areusch

* Remove reference to RFC#0042

Since we decided against RFC#0042, the reference is no longer needed
here.

* Updated with examples of returned loop iterators.

Also, removed implementation details that didn't match the
implementation.  A new stage wasn't necessary, and the loop iterators
could instead be updated in the existing stage, similar to `fuse()`
and `split()`.

* Updated examples of buffer flattening to have valid shape/indices.

Initial example used a buffer shape of `[2,3]`, which was smaller than
the indices used in the example.

* Added discussion on buffer index conventions.

* Typo, forgot to end a sentence.

* Typo fixup, plus integrating in the conclusions from comments.