Improved CI testing for kernels

[devinamatthews]

There are some important facets of the various kernels which are not currently tested:

• [ ] The case of beta == 0 in gemm ukrs. We should pre-populate C with NaNs to make sure it gets overwritten.
• [ ] gemmsup kernels. I don't know how to best test these, but we should make sure that we get complete code coverage (all transa/b cases plus different sizes?). This should also include beta == 0.
• [ ] The case of alpha != 1 in packing kernels. This could use the gemm ukr test code almost as-is.
• [ ] (After merging a future PR which puts edge-case handling in the gemm ukr) Edge cases in the gemm ukr.