Karl Higley

This one is tricky, because Triton [doesn't have a datetime dtype](https://github.com/triton-inference-server/server/blob/main/docs/user_guide/model_configuration.md#datatypes). It could be cast to an integer first, but that would currently need to happen in the Workflow

@Nolski Known issue: https://github.com/OpenMined/syft-proto/issues/91

Is the new action/job added here specifically for the RAPIDS runner? If so, it should probably say that in the name and file name.

There's a (very small) start on this here to cover a bug we ran into: https://github.com/NVIDIA-Merlin/core/blob/main/tests/unit/core/test_dispatch.py

@rogerioxavier It seems like you already have a solution for this issue, which sounds right to me. I'd be happy to review/merge a PR if you'd be willing to open...

Historically we ran them on CPU only in order to avoid creating logjams on the GPU resources when the tests run in Core. This might be a reasonable and/or necessary...

Looks good, needs a test

If the column is truly no longer categorical, then one option here is tag to the column with `Tags.CONTINUOUS` (which will remove `Tags.CATEGORICAL` since no column can be both.) Is...

This looks good! My only comment here is that I wonder if this function might belong in the `io` package, which is where most of the Dask-related functionality currently lives.

Probably fine to add it there for now then. I would like to organize the utils into other more clearly named sub-packages, but that's out of scope for this issue.