Add input/outputShape information to all KDocs of layers

[zaleslaw]

We need to have better documentation for our layers.
A few layers have inputShape/outputShape information in the KDocs, copied from Keras documentation, for example
GlobalAvgPool2D

Please add the same documentation section to all the following layers:

• ELU
• LeakyReLU
• PReLU
• ReLU
• Softmax
• ThresholdedReLU
• DepthwiseConv2D
• SeparableConv2D
• Dense
• Input
• BatchNorm
• AvgPool1D
• AvgPool2D (should be in one format)
• AvgPool3D
• GlobalMaxPool1D
• GlobalMaxPool2D
• GlobalMaxPool3D
• MaxPool1D
• MaxPool2D
• MaxPool3D
• Flatten
• ZeroPadding1D
• ZeroPadding2D
• ZeroPadding3D

[dosier]

Hey, I'd like to contribute a bit again. Could do this one.

One question I have, in some cases Keras specifies different shapes depending upon the data_format being channels_first or channels_last. I do see we have this implemented for the ZeroPadding.. layers but not for any others.

From what I understand the default implementation uses the channels_last format.

There are two conventions for shapes of images tensors: the channels-last convention (used by TensorFlow) > and the channels-first convention (used by Theano).

Deep Learning with Python - François Chollet
(https://stackoverflow.com/a/54967450/12089702)

So I guess for now we just omit the part about the data_format and only take the channels_last part?

[dosier]

Also I guess Input and Flatten do not need this documentation?