CUDA 13: use releases, not nightlies, for PyTorch wheels (#5356)

Contributes to rapidsai/build-planning#208

Follow-up to #5236

Similar to rapidsai/cugraph-gnn#355

PyTorch now has official releases with CUDA 13 support. This switches from nightlies to those releases for wheels in CI here.

Authors:
  - James Lamb (https://github.com/jameslamb)

Approvers:
  - Bradley Dice (https://github.com/bdice)

URL: #5356

Author: jameslamb

dependencies.yaml

@@ -461,9 +461,7 @@ dependencies:
           - matrix:
               cuda: "13.*"
             packages:
-              # TODO: switch from nightlies to releases when there are pytorch CUDA 13 released packages
-              #  ref: https://github.com/pytorch/pytorch/issues/159779
-              - --extra-index-url=https://download.pytorch.org/whl/nightly/cu130
+              - --extra-index-url=https://download.pytorch.org/whl/cu130
           - {matrix: null, packages: null}
       - output_types: [requirements, pyproject]
         matrices:
@@ -475,9 +473,7 @@ dependencies:
           - matrix:
               cuda: "13.*"
             packages:
-              # TODO: remove this '.dev0' (which allows nightlies) once there are pytorch CUDA 13 released packages
-              #  ref: https://github.com/pytorch/pytorch/issues/159779
-              - &pytorch_pip torch>=2.9.0.dev0
+              - &pytorch_pip torch>=2.9.0
               - &tensordict tensordict>=0.1.2
           - {matrix: null, packages: [*pytorch_pip, *tensordict]}
   depends_on_libcudf: