Add divergence module (KL + Jensen-Shannon) for distribution comparison

[Par-python]

Adds entroscope.divergence — distribution-vs-distribution measures, the
deferred sibling of the transfer-entropy work. Answers the data-drift use case:
compare two batches of values (training vs production, period A vs B) and
quantify how far apart their distributions are.

API

divergence.kl(p, q, bins=10)   # Kullback-Leibler, directional, bits
divergence.js(p, q, bins=10)   # Jensen-Shannon, symmetric, bounded [0,1] bits
divergence.plot(p, q, bins=10) # overlay the two binned distributions