[ROCm][CI] Switch ROCm CI runner label from gfx942 to ecosystem.mi350

[akashveramd]

This PR replaces gfx942 label with ecosystem.mi350 label.

[pytorch-bot]

Workflows were awaiting approval. CI has now been triggered for the ciflow labels on this PR.

[scotts]

@akashveramd, why are you making this change? @jithunnair-amd, is this appropriate?

[akashveramd]

@akashveramd, why are you making this change? @jithunnair-amd, is this appropriate?

@scotts: There was an ask from Meta to transition from linux.rocm.gpu.gfx942.1 label as it's a retired runner label.
This is the message we received on slack channel discussion-

Ivan Zaitsev [10:08 PM]
hey folks, our agent flagged that some pytorch-adjacent repos are still pinning retired (?) runner label: linux.rocm.gpu.gfx942.1

1  pytorch/FBGEMM  .github/workflows/fbgemm_gpu_ci_rocm.yml:164  
2  pytorch/FBGEMM  .github/workflows/fbgemm_gpu_pip.yml:158  
3  pytorch/FBGEMM  .github/workflows/fbgemm_gpu_benchmark_rocm.yml:138  
4  pytorch/kineto  .github/workflows/linux_rocm_kineto.yml:23  
5  pytorch/kineto  .github/workflows/linux_rocm_pytorch.yml:23  
6  pytorch/ao  .github/workflows/regression_test_pt2e_gpu.yml:89  
7  pytorch/vision  .github/workflows/tests.yml:57

[jithunnair-amd]

@scotts The ROCm job scheduled on the MI350 runner but timed out when building PyTorch. I took a look at the history of the workflow/job and it last succeeded on Jun 24. That means there's a different issue that needs to be resolved, regardless of runner label. Most likely it's to do with making sure the PyTorch build from source completes within the allotted time, which probably needs some tweaking from a caching perspective or GFX archs etc. @akashveramd can look into that more.

Meanwhile, can we please merge this PR so we can remove dependency on the MI300 runners that we are capacity-constrained on?