fix weight in-place update

[yuxuan-z19]

An in-place update of self.weight in both EvolveGCNH and EvolveGCNO raises a RuntimeError, as shown in the screenshot below. This occurs while using Python 3.10.15 along with torch-2.4.0+cu121 and pyg 2.6.1, with pytorch_geometric_temporal installed from the source.

class EvolveGCNH(torch.nn.Module):
    def forward(self, X, edge_index, edge_weight):
         ...
         if self.weight is None:
            _, self.weight = self.recurrent_layer(X_tilde, self.initial_weight)
        else:
            _, self.weight = self.recurrent_layer(X_tilde, self.weight)
        ...

Inspired by the GCLSTM example, I removed self.weight and made it an argument.

[yuxuan-z19]

I personally, by the way, recommend setting the num_of_nodes and in_channel arguments as follows in the examples:

train_dataset, test_dataset = split_dataset(loader, train_ratio=0.9)
# Note: EvolveGCNH(num_of_nodes, in_channel)
model = EvolveGCNH(train_dataset[0].num_nodes, train_dataset[0].num_features)

This should be helpful for those who are new to this package.