implement oneMKL row-major -> cublas mapping #588
Description
After #586 the cublas -> oneMKL mappings are almost as complete as they can be according to the current oneMKL spec; although note that there are cuda cases that aren't exposed via oneMKL, see e.g. #562. Also some low precision types are not implemented yet: #506
However the oneMKL -> cublas mapping is not complete since row_major cases are not implemented.
cublas uses col_major, but takes a transpose bool. oneMKL also exposes a row_major option. rocblas is the same situation, and this patch:
mapped oneMKL row_major to rocblas implementation via clever usage of transposition.
The cublas backend could do the same, probably via mainly just copy pasting the row_major blas implementations from the rocblas backend.