Transposed Convolution

[notiho]

Add a transposed convolution type to VGSL. In order to match the output size in parallel layers, I added a parameter to the forward pass that describes the expected output size of the layer. (This is maybe a bit awkward and also does not work correctly if the transposed convolution is not the very last layer before the parallel block ends)

[mittagessen]

Thanks. Could you also update the VGSL docs in docs/vgsl.rst? Afterwards I'd merge it.

[notiho]

Sure, I updated the documentation and also added a parameter to control dilation

[mittagessen]

Thanks.