freeze_transform does not work for datasets with child AvalancheDataset in _datasets

[HeLehm]

🐛 Description
If an AvalancheDataset (parent) contains other AvalancheDatasets (children) (in _datasets) and .freeze_transforms() is being called on the parent, the children's transforms are not frozen.

🐜 To Reproduce

def test_avalanche_inherit_groups_freeze_transforms(self):
            original_dataset = MNIST(
                root=default_dataset_location("mnist"), download=True
            )
    
            transform_groups = dict(
                train=(RandomCrop(16), None), eval=(None, None)
            )
            dataset = make_classification_dataset(
                original_dataset, transform_groups=transform_groups
            )
    
            dataset_inherit = make_classification_dataset(dataset)
            x, *_ = dataset_inherit[0]
    
            dataset_frozen = dataset_inherit.freeze_transforms()
            x2, *_ = dataset_frozen[0]
    
            dataset_frozen_reset = dataset_frozen.replace_current_transform_group(
                None
            )
            x3, *_ = dataset_frozen_reset[0]
    
            self.assertEqual(x.size, (16, 16))
            self.assertEqual(x2.size, (16, 16))
            # should be (16, 16) here as RandomCrop(16) should be frozen
            self.assertEqual(x3.size, (16, 16))

🐝 Expected behavior
I would expect the transforms of the children to also be frozen.
In the example x3 should be an Image of size (16, 16), but it is of size (28, 28), the original MNIST image size.

🦋 Additional context
I am working on a fix right now.