Reconsider and optimize CI jobs for OpenCL and SYCL
Summary
We have a mix of CI jobs on AMD and Intel GPUs, from the times when we had to use dedicated machines for them. This is complicated and does not fully reflect the level of support / priority for each platform thus tested.
Now, we have GPUs from all three vendors in our main k8s cluster, so we can safely run them pre-merge and post-merge.