Speed up segmentation and feature calculation

[tulerpetontidae]

The current implementation of feature calculation is extremely slow (takes more time than the segmentation itself for large images). I suggest speeding it up using sparse matrices.

• Post hoc feature calculation on the whole image (FastNucleiFeatures class)
• Add option for sparse matrix storage (convenient for downstream analysis)
• Try to speed up build in NucleiFeatures class. It should be less memory intense but might not add that much speed at the end
• Change example notebook

Happy to hear your suggestions @yozhikoff