Update self-host CI tasks.

[ye-luo]

Our self hosted CI needs some updates.

• sulfur CI Intel21-MPI-CUDA-AFQMC sick due to CUDA 12 Intel Classic incompatibility
• nitrogen CI RadeonVII-ROCm-NoMPI-CUDA2HIP Working
• sulfur CI Clang15-MPI-CUDA-AFQMC-Offload Working
• sulfur CI GCC9-NoMPI-MKL Working
• nitrogen CI GCC9-MPI-CUDA-AFQMC inactive

CPU testing is fast. NVIDIA GPU testing is OK. AMD GPU testing is very slow.

We are missing CPU only + AFQMC. I'm afraid there is breakage already. NVIDIA clang only offload + no CUDA NVIDIA clang only CUDA + no offload AMD ROCm Offload + CUDA2HIP.

My suggestion

• sulfur CI GCC9-NoMPI-MKL basic sanity check the node
• sulfur CI Intel21-MPI-AFQMC Intel classic CPU only
• sulfur CI OneAPI21-MPI-AFQMC Intel oneAPI CPU only #4326
• nitrogen CI RadeonVII-ROCm-MPI CPU only
• nitrogen CI RadeonVII-ROCm-NoMPI-CUDA2HIP AMD HIP + no Offload
• nitrogen CI RadeonVII-ROCm-NoMPI-Offload-CUDA2HIP AMD Offload + HIP
• sulfur CI Clang15-MPI-AFQMC CPU only
• sulfur CI Clang15-MPI-Offload-CUDA-AFQMC NV Offload + CUDA
• sulfur CI Clang15-MPI-Offload-AFQMC NV Offload + no CUDA

In general, GPU builds for each vendor needs 4 variants +/- CUDA/HIP and +/- Offload. We probably won't do all combinations. + CUDA/HIP and + Offload must be covered. In partial variants, I chose ROCm with the HIP only + no offload build and Clang15 with the offload + no CUDA build

[williamfgc]

thanks @ye-luo , adding @PhilipFackler to also get him familiar with our CI

[prckent]

Long since addressed. Lets make a new issue for any CI requests.