bump chgnet to support numpy 2, skip failing matgl test for torch 2.4.1

[DanielYang59]

Summary

• Bump chgnet to support Numpy2

Conflicting torch version

• chgnet 0.4.0 requires torch > 2.4.1 to be NumPy2 compatible Support NumPy 2 CederGroupHub/chgnet#202 (comment)
• matgl requires torch <= 2.2.1 for dgl to work correctly bump chgnet to support numpy 2, skip failing matgl test for torch 2.4.1 #4101 (comment)

Should we skip two failing tests for matgl here as I assume this should be fixed from the dgl side?
Input appreciated.

[DanielYang59]

The unit test failure:

E FileNotFoundError: Cannot find DGL C++ graphbolt library at /home/runner/micromamba/envs/pmg/lib/python3.12/site-packages/dgl/graphbolt/libgraphbolt_pytorch_2.4.1.so

Seem to be related to DGL doesn't have that particular shared object for torch 2.4.1:

~/opt/miniconda3/envs/pymatgen/lib/python3.12/site-packages/dgl/graphbolt$ ls libgraphbolt_pytorch_2.*
libgraphbolt_pytorch_2.0.0.so  libgraphbolt_pytorch_2.1.0.so  libgraphbolt_pytorch_2.1.2.so  libgraphbolt_pytorch_2.2.1.so
libgraphbolt_pytorch_2.0.1.so  libgraphbolt_pytorch_2.1.1.so  libgraphbolt_pytorch_2.2.0.so

should be related to dmlc/dgl#7247 and materialsvirtuallab/matgl#296

dgl==2.1.0
matgl==1.1.3

[shyuep]

Since chgnet is already implemented in matgl, I would rather we unify support for everything that matgl supports using matgl.

[DanielYang59]

Great to know, thanks! But I don't any experience with matgl or chgnet, perhaps need someone with more expertise on this to migrate the implementation