Are tails of gamma prior on tau^2 in decov prior too long?

[jgabry]

Had a conversation with @lauken13 and for e.g., logistic regression or other models where we may not want a long-tailed prior on the sigmas for the "random effects" it could be nice to allow a half-normal or something like that instead of requiring a gamma distribution on tau^2 in the decov prior.