Pysyft: Sigmoid in SMPC failing for some shapes

[LaRiffle]

Describe the bug
Examples to illustrate:

Working

a = torch.Tensor([1,2,3])
a = a.fix_precision().share(bob, alice, crypto_provider = crypto_provider, requires_grad = True)
torch.nn.Sigmoid()(a)

Not working

a = torch.Tensor([[1,2,3],[4,5,6]])
a = a.fix_precision().share(bob, alice, crypto_provider = crypto_provider, requires_grad = True)
torch.nn.Sigmoid()(a)

Error: AssertionError: Must be batches of square matrices

(same with requires_grad=False)

The error seems linked to tensor shape

[AlanAboudib]

It worked before by the way. Like a month ago

[karlhigley]

Might need to add a test for this.

[gmuraru]

I can take it @LaRiffle