Make it easier to share workflow steps with other repositories

[ScottTodd]

Forking this issue from https://github.com/nod-ai/SHARK-TestSuite/issues/288.

This workflow https://github.com/iree-org/iree-test-suites/blob/main/.github/workflows/test_onnx_ops.yml has some duplicated code at https://github.com/iree-org/iree/blob/main/.github/workflows/pkgci_test_onnx.yml.

Most of the duplication is boilerplate and keeping it synced across commits is a bit of a chore. See https://docs.github.com/en/actions/using-workflows/avoiding-duplication for the options available to us.

I'm imagining something downstream like

jobs:
  test_onnx:
    steps:
      - name: Check out external TestSuite repository
        uses: actions/checkout@v4.1.7
        with:
          repository: nod-ai/SHARK-TestSuite
          ref: v1
          path: SHARK-TestSuite
          submodules: false
          lfs: false
      - name: Run onnx tests
         uses: SHARK-TestSuite/.github/workflows/test_onnx.yml
         with:
           pytest_flags: ...
           config_file: ...

or

test_onnx:
  uses: iree-org/test-suite@v1  # or nod-ai/iree-test-suite
  with:
    pytest_flags: ...
    config_file: ...    

where the workflow would include running tests, checking for diffs in the config file and uploading the new file(s), etc.

---

New thoughts since filing that:

For local and CI usage, an entry point script (could be pytest) that runs all test suites here would be helpful.

Some things that change from run to run or machine to machine:

• Available and/or selected hardware and APIs to use (GPUs connected / visible / requseted, use Vulkan, use CUDA, use ROCm, use CPU, etc.
• Test expectations (which should be passing vs which are known failing, are deviations from that okay or should they fail the run?)
• How to shard, parallelize, log, etc.