[release/2.13] Pin test-infra-ref to release/2.13 in validation matrix generation (#8171)

The per-platform validate workflows call
generate_binary_build_matrix.yml without passing test-infra-ref, so it
falls back to the workflow default of 'main'. The build matrix (and its
stable_version / CURRENT_CANDIDATE_VERSION) is therefore computed from
main's generate_binary_build_matrix.py even when validating
release/2.13.

main's CURRENT_CANDIDATE_VERSION does not match the version served on
the test channel for the 2.13 release, so validation installs the wrong
version and smoke_test.py fails with a "Torch version mismatch, expected
<x> for channel test" error.

Pin test-infra-ref: release/2.13 in the matrix-generation calls so the
matrix is computed from the release branch.

Authored with the assistance of Claude Code.

Author: atalman

.github/workflows/validate-aarch64-linux-binaries.yml

@@ -96,6 +96,7 @@ jobs:
       package-type: wheel
       os: linux-aarch64
       channel: ${{ inputs.channel }}
+      test-infra-ref: release/2.13
       with-cuda: enable
       with-cpu: enable
       use-only-dl-pytorch-org: ${{ inputs.use-only-dl-pytorch-org }}

.github/workflows/validate-linux-binaries.yml

@@ -106,6 +106,7 @@ jobs:
       package-type: wheel,libtorch
       os: linux
       channel: ${{ inputs.channel }}
+      test-infra-ref: release/2.13
       use-only-dl-pytorch-org: ${{ inputs.use-only-dl-pytorch-org }}
       with-xpu: enable
 

.github/workflows/validate-macos-arm64-binaries.yml

@@ -96,6 +96,7 @@ jobs:
       package-type: wheel,libtorch #  We stopped producing conda nightlies
       os: macos-arm64
       channel: ${{ inputs.channel }}
+      test-infra-ref: release/2.13
       use-only-dl-pytorch-org: ${{ inputs.use-only-dl-pytorch-org }}
 
   macos-arm64:

.github/workflows/validate-windows-binaries.yml

@@ -96,6 +96,7 @@ jobs:
       package-type: wheel,libtorch #  We stopped producing conda nightlies
       os: windows
       channel: ${{ inputs.channel }}
+      test-infra-ref: release/2.13
       use-only-dl-pytorch-org: ${{ inputs.use-only-dl-pytorch-org }}
       with-xpu: enable