Add HIP TRSM

[abouteiller]

This is based over #121 as this is needed for it to work

I have used the info handles in a different way than we did for potrf, I think its better than having to set taskpool internals in the wrapper

[QingleiCao]

Some CUDA and HIP parameters are separate. Do we have a plan to make CUDA and HIP supported at the same time in the future?

BTW, what I'm doing in HiCMA is adding a header file to redefine CUDA variables (at least for most of them) so that minimal change is needed.

[abouteiller]

Some CUDA and HIP parameters are separate. Do we have a plan to make CUDA and HIP supported at the same time in the future?

Yes eventually that would be ideal. Right now it is not possible, but the code in dplasma is written in the way that would make it possible when parsec supports it.

BTW, what I'm doing in HiCMA is adding a header file to redefine CUDA variables (at least for most of them) so that minimal change is needed.

We decided against doing a .h trickery because that almost works, but not always (rocsolver and cusolver do not always take the same parameters, in some cases you need a workspaces, but not in the other, etc.), and having a common code is therefore problematic and more error prone than helpful.

[devreal]

LGTM

[abouteiller]

CI error is #115 and can be ignored