Make cu130 the primary pip index, keep torch pin bare

ZeroGPU's config validator rejects torch==2.9.1+cu130 (only bare
2.8.0/2.9.1/2.10.0/2.11.0 allowed), but PyPI's bare 2.9.1 is a cu128
build and --extra-index-url won't outrank PyPI for the same base
version. Flip priority: cu130 index primary, PyPI extra. pip resolves
the bare torch==2.9.1 to the cu130 wheel and the validator passes.

Co-Authored-By: Claude Opus 4.8 (1M context) <noreply@anthropic.com>

requirements.txt

@@ -1,10 +1,13 @@
-# ZeroGPU build container ships the CUDA 13.0 toolkit; pull torch/vision from the
-# cu130 index so diff-gaussian-rasterization compiles against a matching CUDA
-# (torch 2.8 has no cu130 wheel, so 2.9.1 is the floor). The explicit +cu130
-# local version forces pip to the cu130 index instead of PyPI's cu128 build.
---extra-index-url https://download.pytorch.org/whl/cu130
-torch==2.9.1+cu130
-torchvision==0.24.1+cu130
+# ZeroGPU build container ships the CUDA 13.0 toolkit, so torch must be a cu130
+# build for diff-gaussian-rasterization to compile (torch 2.8 has no cu130 wheel,
+# so 2.9.1 is the floor). ZeroGPU's config validator rejects a +cu130 local
+# version on the torch pin, and an --extra-index-url doesn't outrank PyPI for the
+# same base version. So make the cu130 channel the PRIMARY index (PyPI demoted to
+# extra): pip resolves the bare torch==2.9.1 to the cu130 wheel, validator stays happy.
+--index-url https://download.pytorch.org/whl/cu130
+--extra-index-url https://pypi.org/simple
+torch==2.9.1
+torchvision==0.24.1
 numpy==1.26.4
 pillow==10.4.0
 diffusers[torch]==0.30.3