A better story for normalization constants #74
Description
Relates to #70. It's very easy to get messed up trying to keep track of what's normalized and what's not when dealing with non-tractable densities in the monte carlo module. I'm wondering if some form of statically typed unit analysis is viable. The idea would be, every density for a distribution has a normalization constant, which in practice is either 1 or some unknown function of the distribution's parameters. As these densities are manipulated, their normalization constants should tag along and combine/cancel as appropriate.