FEAT: Constrat Limited (Adaptive) Histogram Equalization

[neutrinoceros]

See section 4.2 from Adaptive histogram equalization and its variations, Pizer et al, 1987)

The procedure of clipping a histogram seems general enough that it could be implemented once for all, however I expect it would require care to make it compatible with SWAHE (#301).

[neutrinoceros]

An important question to answer first is how should this be parametrized ?
I'm inclined to imitate skimage.exposure.equalize_adapthist which accepts values between 0 and 1, with a couple caveats:

• I don't want any clipping enabled by default (this seems to imply the default clipping threshold to 0.0 ?)
• the exact normalization they use isn't clearly documented. The only one that seems to make sense to me would be that 1.0 correspond to the total number of pixels being binned (so the area of the histogram, before renormalization, would tend to 0 as we approach 1.0, hence ==1.0 wouldn't be allowed.), but this needs clarification
• I also want to allow users to specify if the clipping/renormalization procedure should be iterative and how to stop it; typically this would be determined either as a maximum number of iterations, or a tolerance margin.

[neutrinoceros]

The only one that seems to make sense to me would be that 1.0 correspond to the total number of pixels being binned

Confirmed: 1.0 is equivalent to the maximum possible bin count, i.e., the total number of pixels in the contextual region (tile/window)

https://github.com/scikit-image/scikit-image/blob/07cf8f8f7d45b0f1742e36cd8185e0912d05b705/src/skimage/exposure/_adapthist.py#L22