log_sum_exp on wrong variable (scalar vs. arrays)

[bob-carpenter]

Using log_p_new instead of the right log_p_news in the log_sum_exp calc for the "right" way to calculate log expected density.

[jgabry]

Also, where it says

a little wrangling with Jensen’s inequality shows that the expectation of the log is less than or equal to the log of the expectation

this is true but the math below this line has it reversed