Add support for custom callbacks in graphgym.train

[DhyeyMavani2003]

Description

This PR implements the feature requested in #10386 to allow users to pass custom PyTorch Lightning callbacks via the trainer_config parameter in torch_geometric.graphgym.train.

Changes

• Modified torch_geometric/graphgym/train.py to extract and extend custom callbacks from trainer_config
• Handles both single callback and list of callbacks
• Maintains backward compatibility with existing code

Usage Example

from torch_geometric.graphgym.train import train

# Pass custom callbacks
trainer_config = {
    'callbacks': [my_custom_callback],
    # other trainer config options...
}

train(model, datamodule, trainer_config=trainer_config)

Testing

• ✅ All pre-commit hooks pass (linting, formatting, etc.)
• ✅ Existing tests remain compatible
• ✅ The implementation follows the exact suggestion from the issue

Fixes #10386