[TKW] Feature Request: waves_per_block dimensionality should match workgroups

[GMNGeoffrey]

The waves_per_block argument to tkw.HardwareConstraint currently has to be of length 3. Wave supports logical workgroup distribution along more than 3 dimensions and then figures out the mapping later. It should do the same with waves_per_block. Related to this, there are several sharp edges with the current setup:

1. If you have a workgroup dimension with block size other than 1 at index 3 or greater, the Wave compiler tries to look up waves_per_block at that index and fails with an IndexError
2. If you try to make waves_per_block longer than 3, the program crashes (looks like a segfault on the C side or something).
3. You have to specify waves_per_block for dimensions that you're not distributing along.

FYI @harsh-nod