SegNet Model + Example

[penguinmenac3]

Is your feature request related to a problem? Please describe.
Segnet is one of the most important segmentation networks and nicely illustrates concepts like upconvolution. Having a clean implementation would benefit lot's of other potentially interesting network implementations.

Describe the solution you'd like

1. Implement SegNet-Model (create_model-function) in starttf.models (using the vgg16_encoder implementation as an encoder)
2. Write a full SegNet example (like the mnist example) where the network is trained on cityscapes dataset.
   2.1. Write a loss.py
   2.2. Write a prepare_training.py
   2.3. Write a hyper_params.json
   2.4. Write a train.py
3. If required add cityscapes to the opendatalake.

Describe alternatives you've considered
None.

Additional context
This might be a good issue for someone who wants to start contributing to starttf.