Always run nvmath-python tests on PRs

Remove the main-branch-only guard that was added in #723. The same PR
already accelerated the tests enough for per-PR runs.

Closes #850

Author: leofang

.github/workflows/ci.yaml

@@ -247,7 +247,6 @@ jobs:
       matrix_filter: map(select(.ARCH == "amd64" and (.CUDA_VER | split(".") | .[0] | tonumber == 12))) | group_by(.CUDA_VER|split(".")|map(tonumber)|.[0]) | map(max_by([(.PY_VER|split(".")|map(tonumber)), (.CUDA_VER|split(".")|map(tonumber))]))
 
   test-thirdparty-nvmath:
-    if: ${{ github.ref_name == 'main' }}
     needs:
       - build-linux-64
       - compute-matrix