[Deps] cherry-pick #25576 onto release/v0.5.12

[Kangyan-Zhou]

Summary

Cherry-pick of #25576 (commit a449ee4) onto release/v0.5.12.

Switches nvidia-cutlass-dsl to the [cu13] extra at version 4.5.1, matching what is already on main. This is needed on CUDA 13 images (e.g. GB300) where the non-[cu13] wheel ships CUDA-12.9 binaries that miss the sm_110 → sm_101 alias and crash FlashAttention 4 / cute kernels at runtime.

Changes

• python/pyproject.toml: nvidia-cutlass-dsl==4.5.0 → nvidia-cutlass-dsl[cu13]==4.5.1

Conflict resolution

The source commit on main was a pure version bump ([cu13]==4.5.0 → [cu13]==4.5.1) because main already had the [cu13] extra from a prior commit. On release/v0.5.12 the [cu13] extra was never picked up, so the line still read nvidia-cutlass-dsl==4.5.0. Took the incoming line so this branch ends up with the same final state as main ([cu13]==4.5.1), which is the intent of the original PR title ("Use cu13 extra for nvidia cutlass dsl").

Test Plan

• Install on a CUDA 13 environment (GB300) and confirm nvidia-cutlass-dsl resolves to 4.5.1 with the cu13 extra, and that FA4 / cute kernels no longer crash on sm_110.

---

CI States

Latest PR Test (Base): ❌ Missing run-ci label -- add it to run CI tests.
Latest PR Test (Extra): ❌ Blocked -- run-ci is required first.

[gemini-code-assist]

Warning

You have reached your daily quota limit. Please wait up to 24 hours and I will start processing your requests again!