Use nplog1p in poisson/negbinom likelihood function

[twallema]

• I have checked to ensure there aren't other open Pull Requests for the same update/change
• I have updated the documentation accordingly.

Describe your fixes/additions/changes

In ll_poisson and ll_negative_binomial, there was a test to see if the ydata or ymodel were smaller than or equal to one. If this was the case, then the minimal model/data point were found and raised to 1e-6. This is,

1. Unnecessarily complex, you can just add one to both ymodel and ydata.
2. May actually introduce bias, because a log of a small number tends to negative infinity.

So now, by default, in ll_poisson and ll_negative_binomial,

ymodel += 1
data += 1

I've verified this gives the same results in the workflow & influenza 17-18 tutorial.