Fix smoke tests

[rogthefrog]

Wins

• Fix missing credentials
• Add Google authentication for Vertex AI
• Temporarily disable a SUT on Huggingface until the HF client can handle its own API's datetime fields 🙄
• Add an option for fake secrets to use specified real secrets in case they are needed for some steps (in this case, downloading a prompt file from modellab, which requires an auth token)

Fails

I tried to extract the credential/secret logic into a dependent workflow so it could be shared between the app test and smoke test workflows. That didn't work because:

• a secrets.toml file created in the runner file system in the dependent workflow doesn't get preserved when the caller workflow runs
• the canonical way to share files between workflows is to upload them as artifacts. I didn't want to do with secrets.toml because it's full of secrets, and I couldn't find enough information about artifact ACLs to be comfortable uploading it as an artifact.
• setting the contents of the secrets.toml file as an output of the dependent job does not work, because github actions helpfully strip out outputs that include secrets.

So I just C&P'ed the secrets section into a second workflow, like a neanderthal.

[github-actions]

MLCommons CLA bot All contributors have signed the MLCommons CLA ✍️ ✅