ci: add NVIDIA workers and NVMf {RDMA,TCP} jobs

[mchomnic]

Rename the nvmf-rdma job to vm-nvmf-rdma for generic rdma-vm workers and add a dedicated nvmf-tcp workflow.

[mhae]

It's not clear to me why we need to have a common workflow. This workflow is specific to HPE's VM and config (i.e. CX6 and actual NVMe drives).
I also don't want to run nvmf tcp on our VM.
I suggest we keep these jobs separate.

[mhae]

@mchomnic , remind me again what HW you are using? Is it ARM or x86?
I wonder if we should have specific runners based on HW vs "common" NVMF/RDMA runners.
What if you want to add some tests that uses a HW feature that HPE's runner doesn't have?
On the other hand, it would help with high availability....
Note, if we change labels, then I have to update and restart my runners.

[mchomnic]

Hi @mhae,
The common workflow was a @tomzawadzki's suggestion as baseline to discussion.
The other options is to stay with separate label, eg: nvidia-rdma-vm (or something similar), but the consers was, do we want to run nvmf-rdma twice in per-patch.
I have x86 VMs with CX9 vfs and emulated nvme.

[mchomnic]

@mhae @tomzawadzki , I changed the approach. I added separated jobs for nvidia's jobs. WDYT?
P.S. Two nvidian's workers are already added to spdk/spdk-ci . I want to check the workflow.

[karlatec]

Can we re-label the runners to have more details or use more then one label? E.g. using [nvidia, x64, ubuntu-2404, vm] provides much more information at first glance.

[karlatec]

@mchomnic how about this one? It's much less effort re-labeling the runners now rather than doing it post-merge when there's builds queued up.

[mchomnic]

Sure, I'm open for any changes. I just want to discuss it before I will update config, because I have already done twice.
What about [Linux, X86, ubuntu-2404, nvidia-vm, gen-vm]? @tomzawadzki