Add transpose routines to BLAS-like extensions

[andrewtbarker]

This PR adds transpose and batch-transpose BLAS-like extensions imatcopy, omatcopy, omatadd, imatcopy_batch, omatcopy_batch, and omatadd_batch to the spec.

[andrewtbarker]

@aaronjohnson I addressed your comments and added the table you suggested, let me know what you think.

[mmeterel]

@andrewtbarker is this good to merge?

[andrewtbarker]

@mmeterel we have a TAB meeting on these changes next week, I think it makes sense to wait until after the meeting to merge.

[andrewtbarker]

@aaronjohnson @mmeterel @lhuot @spencerpatty I've modified the spec based on feedback from TAB and others:

1. Added documentation of in-place use of omatadd to do C += alpha A and other previously unsupported cases;
2. ~Added geam as a simple wrapper around omatadd;~
3. Added omatcopy2 because it was missing from spec;
4. More uniform references to strided-batch and group-batch for all the APIs.

I'd appreciate any comments or suggestions.

[andrewtbarker]

In my view this is ready to merge - let me know if you have any concerns.

[andrewtbarker]

@spencerpatty I made most of the changes you suggested (add "dense" and use "specify" instead of "implement") in several places. Regarding transpose options: I clarified the transpose datatype description here, which is referenced multiple times from the spec, but I left the individual pages because their description of the transpose parameters is consistent with elsewhere in the spec (ie blas::gemm).