Conv 1d and unet refactoring

[DerThorsten]

This PR adds the following:

Convolutions:

• 1D convenient classes for all convolutions
• ND convenient classes for all convolutions ( eg ConvELU where dim is a keyword of the constructor, only 1D, 2D, and 3D works)

UNetBase:

• major refactoring of UNetBase:
  • added start block and end block, this allows for an easier implementation of UNet. The start block can be used to increase the features from in_channels to initial_features before the actual unet, the end block is used to convert the output features of the unet to the desired number of channels (@constantinpape what do you think about theses changes)
  • cleaned code

This is still a work in progress, the ResUnet still needs the refactoring, also the examples are not yet up-to date

[constantinpape]

* The start block can be used to increase the features from `in_channels` to `initial_features` before the actual unet, the end block is used to convert the output features of the unet to the desired number of channels (@constantinpape what do you think about theses changes)

I like the idea. Didn't have much time to look at the actual changes, will take a closer look on Monday.