Add distributed-ucxx to dependencies.yaml

[rjzamora]

Adds distributed-ucxx to the depends_on_dask_cuda section of dependencies.yaml, since rapidsmpf technically requires distributed-ucxx (or ucxx + ucx-py) to function on a dask-cuda cluster.

This change should make it a bit easier to build multi-gpu Polars with rapidsmpf.

[TomAugspurger]

I'm pretty stumped by the error message here:

ERROR: Cannot install rapidsmpf-cu12[test]==25.10.0a56 because these package versions have conflicting dependencies.

The conflict is caused by:
    rapidsmpf-cu12[test] 25.10.0a56 depends on distributed-ucxx==0.46.* and >=0.0.0a0; extra == "test"
    The user requested (constraint) distributed-ucxx==0.46.*,>=0.0.0a0

xref #339

[pentschev]

I'm pretty stumped by the error message here:

ERROR: Cannot install rapidsmpf-cu12[test]==25.10.0a56 because these package versions have conflicting dependencies.

The conflict is caused by:
    rapidsmpf-cu12[test] 25.10.0a56 depends on distributed-ucxx==0.46.* and >=0.0.0a0; extra == "test"
    The user requested (constraint) distributed-ucxx==0.46.*,>=0.0.0a0

xref #339

That's because the changes here are insufficient, this is what I wrote in a Slack conversation with what needs to be done:

Yeah, you’ll have to change many more things. I think the easier is just changing all ucxx entries (not libucxx) to distributed-ucxx. You’ll then need to change the conda recipes too, I now see we have ucxx in https://github.com/rapidsai/rapidsmpf/blob/branch-25.10/conda/recipes/librapidsmpf/recipe.yaml, but that is wrong, it should be only libucxx in the C++ package, and then add distributed-ucxx to https://github.com/rapidsai/rapidsmpf/blob/branch-25.10/conda/recipes/rapidsmpf/recipe.yaml. The ucxx package alone is not necessary anywhere, C++ packages only need libucxx and Python packages should only need ucxx.