[ROCm][CI] First draft of ROCm build workflow

[jithunnair-amd]

This PR introduces a new build workflow for ROCm

• Uses ROCm6.4 PyTorch 2.9.1 wheels to match the bare metal environment on ROCm runners (will switch to PyTorch nightly wheels in a follow-up PR)
• ROCm6.4 almalinux image has devtoolset-14, so need the PATH setting for that since rust needs a new enough compiler
• Set USE_TENSOR_ENGINE=0 to disable rust errors when building cuda-sys, nccl-sys etc.
• The eventual goal is to invoke this from ci.yml, but I added a workflow_dispatch trigger to allow us to pipeclean this workflow first by manually triggering it (this will need granting extra permissions though). Alternately, if we enable the ciflow/rocm label-based trigger mechanism for this repo (similar to how we do on pytorch/pytorch), we can test it on the PR itself (a pull_request trigger will not work to test on the PR because the AWS OIDC step will fail on a forked repo branch).

Depends on #2000 and #2032

[jithunnair-amd]

cc @mreso for awareness

[jithunnair-amd]

@pytorchbot label "ciflow/rocm"

[pytorch-bot]

To add these label(s) (ciflow/rocm) to the PR, please first approve the workflows that are awaiting approval (scroll to the bottom of this page).

This helps ensure we don't trigger CI on this PR until it is actually authorized to do so. Please ping one of the reviewers if you do not have access to approve and run workflows.

[jithunnair-amd]

@pytorchbot label "ciflow/rocm"

[pytorch-bot]

To add these label(s) (ciflow/rocm) to the PR, please first approve the workflows that are awaiting approval (scroll to the bottom of this page).

This helps ensure we don't trigger CI on this PR until it is actually authorized to do so. Please ping one of the reviewers if you do not have access to approve and run workflows.

[jithunnair-amd]

@pytorchbot label "ciflow/rocm"

[pytorch-bot]

To add these label(s) (ciflow/rocm) to the PR, please first approve the workflows that are awaiting approval (scroll to the bottom of this page).

This helps ensure we don't trigger CI on this PR until it is actually authorized to do so. Please ping one of the reviewers if you do not have access to approve and run workflows.

[huydhn]

@jithunnair-amd I can see the rocm tag being pushed to https://github.com/meta-pytorch/monarch/tree/ciflow/rocm/2001, so the bot is working as expected. The reason why it doesn't show up here in your PR is that you need to land the PR first. For this PR, you can just add pull_request trigger to run it, then remove the trigger before landing