Document weight splitting in symbolic.optimal_weights()

[ketch]

Calling

f=symbolic.optimal_weights(k,(0.2,))
f[0]

gives

{'k': 3,
 'n': 1,
 (0, 0): (0.144615384615376, 0.0723076923076879),
 (0, 1): (0.152307692307692, 0.304615384615384),
 (0, 2): (2.16000000000000, 1.08000000000000)}

which was confusing to me, since each of the dictionary values should be a number, not a tuple. I only realized what was going on after inspecting the code; it automatically gives splittings when weights become negative. It would be nice to document this and show how the return values should be used.