Ignoring / padding of image boundary nodes

[marjanfamili]

Question : Should we (A) discard the nodes at the edges of image right from the start, or (B) keep the edges but not include them in training.

Points in favour of (A) :
1- The nodes that are at the edges are a small proportion of the total number of edges and annotated edges, discarding them does not affect the training / inference
2- if the edges are important bounding-box and keep_patch_fraction should be chosen accordingly

Points in favour of (B):
1- The edges can be quite important at the time of inference and therefore we can use padding to make the size of patches identical

if we have used augmentation with padding during training, the network will not recognise hard edges as features

TODO: needs to fix #134

[marjanfamili]

selected (B) : For now we have decided to keep the edges and pads the patches.