Parameterize infection feedback strength in terms of the p_ed_visit

[dylanhmorris]

Since infections are not directly observed, there's strong posterior correlation between the strength of the infection feedback and the inferred p_ed_visit given infection. Adding additional signals may weaken this problem, but it still seems best to parameterize the infection feedback relative to the infection --> observable rate to break the posterior correlation altogether. If there's concern about interpretability, one can always still place the prior and/or report the posterior on the rate in terms of raw incident infections.

[dylanhmorris]

had guessed that this was the case but confirmed it with the analyses in #164