could clustering use label() from scipy.ndimage instead of scikit-image?

[Thomas Gilgenast]

there is a function for labeling connected components in scipy:

https://docs.scipy.org/doc/scipy-0.16.0/reference/generated/scipy.ndimage.measurements.label.html

this would allow us to drop an optional dep (scikit-image) which must be installed via wheel on windows iirc

this would also allow us to keep only one version of each clustering function, as opposed to the current system which has two separate versions for each function

[Thomas Gilgenast]

fixes #34

removed scikit-image - no longer referenced in install docs - no longer used in two_way_thresholding() - removed from extras_require

removed _slow() versions of cluster counting and size thresholding functions - "fast" versions use scipy and are guaranteed to import correctly

cluster counting step now skips '' and 'background' clusters - these will no longer show up in the output of count_clusters()

threshold tool now always prints cluster counts - it is not possible to disable cluster counting via the tool - it can only be done via a kwarg on the function that is not exposed in the command line tool

→ <<cset d0859af812d2>>