Bug fix and add additional tests for Dataset and DataModule

[FilippoOlivo]

In this PR, I fix a bug in get_all_data function in Dataset. The code for reproducing the issue is the following

input=torch.stack([torch.zeros((1,))+i for i in range(1000)])
target=input
problem = SupervisedProblem(input,target)
datamodule = PinaDataModule(problem,
        train_size=0.7,
        test_size=0.2,
        val_size=0.1,
        batch_size=64,
        shuffle=False,
        repeat=False,
        automatic_batching=None,
        num_workers=0,
        pin_memory=False,)
datamodule.setup('fit')
datamodule.setup('test')
print()
print('Using batch_size 64')
print(f'{(len(datamodule.train_dataset)==700)=}')
print(f'{(len(datamodule.test_dataset)==200)=}')
print(f'{(len(datamodule.val_dataset)==100)=}')
print()
print(f'{(len(datamodule.train_dataset.get_all_data()["data"]["input"])==700)=}')  # <==== False but must be True
print(f'{(len(datamodule.test_dataset.get_all_data()["data"]["input"])==200)=}')   # <==== False but must be True
print(f'{(len(datamodule.val_dataset.get_all_data()["data"]["input"])==100)=}')    # <==== False but must be True

Moreover, I remove a useless shuffle operation in PinaDataModule when creating the Dataloader (shuffle has already been performed). Finally, I improve the test of both dataset and datamodule classes.