Divergences with unused variables

[cluhmann]

@jessegrabowski

I’ve been defaulting to Gamma(2, x) for strictly positive variables for a while now

I have also switched to this. The idea of nearly-zero scale parameters doesn’t seem sensible as a general rule.