CUDA 13: Depend on CUDA builds of `xgboost` / `rapids-xgboost` for `rapids` packages

[jameslamb]

Description

As of #795, there were not yet any XGBoost CUDA 13 packages... so the first CUDA 13 rapids nightly packages shipped with a dependency on xgboost CPU-only builds.

Once there are xgboost packages, that workaround should be reverted and rapids should again depend unconditionally on rapids-xgboost.

Benefits of this work

• sets consistent expectations for what rapids installs, regardless of CPU architecture or CUDA version
• improves peformance of XGBoost workloads using CUDA 13 conda packages

Acceptance Criteria

• rapids depends unconditionally on rapids-xgboost

Approach

Get xgboost CUDA 13 packages up:

• conda packages on conda-forge (Add CUDA 13 builds conda-forge/xgboost-feedstock#253)
• conda packages on rapidsai / rapidsai-nightly ( Add CUDA 13 builds xgboost-feedstock#100 )

Then revert the workaround added in #795

Notes

N/A