[CI] chore: support cuda 13.0.2

[stmatengss]

Description

Type of Change

• Types
  • Bug fix
  • New feature
    • Transfer Engine
    • Mooncake Store
    • Mooncake EP
    • Integration
    • P2P Store
    • Python Wheel
  • Breaking change
  • CI/CD
  • Documentation update
  • Other

How Has This Been Tested?

Checklist

• I have performed a self-review of my own code.
• I have updated the documentation.
• I have added tests to prove my changes are effective.

[gemini-code-assist]

Note

Gemini is unable to generate a summary for this pull request due to the file types involved not being currently supported.

[stmatengss]

@UNIDY2002 I can provide some failed logs for these test cases.

[stmatengss]

  File "/sgl-workspace/sglang/python/sglang/srt/mem_cache/storage/mooncake_store/mooncake_store.py", line 160, in __init__
    from mooncake.store import MooncakeDistributedStore
ImportError: libcudart.so.13: cannot open shared object file: No such file or directory

Wheels built with CUDA 13.0 may not be compatible with lower CUDA versions.

[UNIDY2002]

Do you think it a good idea to publish separate packages, like mooncake-transfer-engine-cu12, mooncake-transfer-engine-cu13?

[stmatengss]

Do you think it a good idea to publish separate packages, like mooncake-transfer-engine-cu12, mooncake-transfer-engine-cu13?

How about your opinion? @ShangmingCai @ykwd It seems cu12 and cu13 are two significantly different environments.

[ShangmingCai]

I think maybe we can build mooncake-transfer-engine-cu13 for CUDA 13, and let the original one serve CUDA 12 as it is doing now.

[ShangmingCai]

Can we learn how PyTorch handles this?

[UNIDY2002]

PyTorch essentially maintains its own registry for various CUDA versions, without depending on pypi.