Add AvgPool1D layer

[mkaze]

Resolves #61. This PR adds support for AvgPool1D layer.

• Implement layer class
• Add unit tests for layer implementation
• Add support for importing models containing this layer from JSON config files
• Add support for exporting models containing this layer to JSON config files

[zaleslaw]

Remove dataFormat and change the params in constructor

[zaleslaw]

Could we add here the default value?

[mkaze]

Since in pooling layers it's very common for strides to be the same as pool size value, the benefit of using null as default value is that the user won't need to set it explicitly and we just default to pool size value instead. This is the same behavior in Keras and PyTorch.

[zaleslaw]

Kotlin and Python users have a different experience with nullability, and Kotlin is trying to be a place without nulls.
As I commented in the MaxPool1D layer PR, please change it to the default value 1,2,1.

We need to keep control of the parameters that could be a hyper-parameter for AutoML tasks and show it to the user.
As a variant - to make a warning if stride and pool are not equal with reminder that it's a common practice

[mkaze]

I see. I understand your point about removing nullability and hidden dependencies. However, I don't get the part about losing control of the parameter during hyper-parameter tuning scenarios or architecture search. Anyways, since I am new to Kotlin I think you have a much better understanding of what would be the best decision here and I would follow it gladly :)

[zaleslaw]

I meant that we train the user to use only one parameter and leave the second null to change together. Perhaps, looking for a new architecture, a method a la NAS, something with this approach will be missed, at least I think so.

[mkaze]

@zaleslaw I just merged it with master branch and updated it with recent changes.

[mkaze]

@zaleslaw I changed the strides type and default value per your request.