handle infinities in (regularized) log-likelihood bar

[jeisner]

See #81 on infinite weights.
How should we display infinities on the LL bar? Right now we just get some random number in there, which is confusing and misleading. We should special-case somehow, either graphically or just by displaying explanatory text in place of the log-likelihood bar. Here are the cases to deal with.

If LL is finite, regularizer may be:

• finite (usual case)
• -Infinity (because C > 0 and some weights are +/- Inf (see allow solver to find -Infinity #81); LL is nonetheless finite because only unobserved events got 0 probability)
• +Infinity (because C < 0 and some weights are +/- Inf)

If LL is -Infinity (because infinite weights resulted in 0 probability for observed events), regularizer may be:

• 0 (if C = 0)
• -Infinity (if C > 0)
• +Infinity (if C < 0)

Note that R(theta) is certainly +Infinity whenever LL is -Infinity, but may be dropped anyway if C=0. (In this case, we treat 0*Infinity as 0.)