Replacing PyHIP with new official python wrapper of ROCm HIP

[MiloLurati]

This pull request replaces PyHIP with the new official python wrapper of rocm HIP and enables passing DeviceArray types (a HIP Python type) as an argument.

[sonarqubecloud]

Quality Gate passed

Issues
10 New issues
0 Accepted issues

Measures
0 Security Hotspots
0.0% Coverage on New Code
0.0% Duplication on New Code

See analysis details on SonarQube Cloud

[benvanwerkhoven]

Hi @MiloLurati,

Thanks for the pull request! I think it's ready, or would you like to leave it in draft for now? The main reason to wait is perhaps that hip-python is not indexed in PyPI yet.

[MiloLurati]

Hi @benvanwerkhoven ,

Yes, I can un-draft it. I've been tuning quite a lot lately with Kernel Tuner and HIP Python and have also been using the DeviceArray wrapper to allocate device memory in the Python script. I haven't encountered any problems so far. If it's okay with you, we could merge these changes.

[stijnh]

I've also add support for half/bfloat16 precision to the HIP backend. I will wait with merging my changes until after the new HIP backend is merged so that I can rebase my changes to use PyHIP.