fix: add runtime_shape argument in batchnorm-related classes to match…

[euphoria0-0]

This PR is related to issue #450 .
BatchNorm1dToQuantScaleBias would provide the same shape: (B, D) -> (B, D).

dim = 100
batch_norm = BatchNorm1dToQuantScaleBias(num_features=dim, runtime_shape=(1, -1))
inputs = torch.rand(size=(5, dim))
outputs = batch_norm(inputs)
print('Outputs:', outputs.shape)

Outputs: torch.Size([5, 100])

batch_norm = BatchNorm1dToQuantScaleBias(num_features=dim, runtime_shape=(1, -1, 1))
inputs = torch.rand(size=(5, dim, 32))
outputs = batch_norm(inputs)
print('Outputs:', outputs.shape)

Outputs: torch.Size([5, 100, 32])

[Giuseppe5]

Thanks, I will take a look at this!