DAGLayer and DAGGather reject glorot_normal/xavier_normal init

Bug

DAGLayer and DAGGather both document init="glorot_normal" and init="xavier_normal" as supported options, but passing either raises ValueError instead of constructing the layer.

To Reproduce

Steps to reproduce the behavior:

1. from deepchem.models.torch_models.layers import DAGLayer
2. DAGLayer(init="glorot_normal")
3. Raises ValueError: Unsupported init: glorot_normal

Same happens with DAGGather(init="glorot_normal"), and with init="xavier_normal" on both classes.

Expected behavior

Layer should construct normally, using nn.init.xavier_normal_ for the weights, same as it already does for glorot_uniform/xavier_uniform.

Environment

• OS: Linux
• Python version: 3.12
• DeepChem version: 2.8.1.dev (current master)
• RDKit version (optional): N/A, not needed to reproduce
• TensorFlow version (optional): N/A, not needed to reproduce
• PyTorch version (optional): 2.x
• Any other relevant information: bug is in _initialize_weights for both classes in deepchem/models/torch_models/layers.py

Additional context

Looks like a comparison bug — self.init (a str) is compared with == against a list instead of checking membership with in, so that branch can never match. Happy to submit a fix if useful.