Adding new sampleable distributions: lognormal and custom

The "custom" sampleable distribution provides samples for any distribution defined by a curve. It is done in two different flavors: a simple algorithm (accurate, adapted to smooth curves) and a method using a kernel density - less accurate but faster and more adapted to noisy curves (thanks to @mrognlie).

Tests are added, to make sure we sample the proper distribution functions.

Also added a "log10" version of a custom distribution (with Kernel density).