Support CUDA 13 scan comparators

[jeongseok-meta]

Summary

• replace the removed cub::Equality scan comparator with thrust::equal_to<int64_t>
• include the public Thrust functional header explicitly

Why

CUDA 13's CUB no longer provides cub::Equality, which prevents the extension from compiling. NerfAcc's scan keys are int64_t in both the forward and reverse-iterator paths, so the typed Thrust comparator preserves the existing behavior.

Using the Thrust functor also keeps the source compatible with older supported CUDA toolkits without a CUDA-version conditional.

Closes #306.

Validation

• Reproduced the unpatched CUDA 13.0.88 failure with eight missing cub::Equality errors.
• Built the full extension with PyTorch 2.10.0 and CUDA 13.0 for an RTX 5080 (sm_120).
• pytest -q tests/test_scan.py: 4 passed.
• pytest -q: 23 passed.
• git diff --check.