Add ResUnet-a and geometric errors

This release adds the following:

• ResUnet-a architecture used for field delineation
• geometrical errors for performance evaluation

New modules and architectures

New features:

• PSE-TAE architecture
• residual convolutional layers
• Pyramidal Pooling Module
• Tanimoto loss
• MatthewCorrelationCoefficient metric

Minor improvements and bug fixes

Conversion to `tf-v2.x`

This version includes code converted to tf-v2.x