Add priors based on event rates within CalibrationMap

Since CalibrationMaps now store things like the segments from which the samples are taken (by managing separate quivers), we can easily estimate rates of glitches and cleans as simply the number divided by the live times. These could then be incorporated in the mapping from loglike to pglitch.

We need to think about exactly what the priors should be as a function of the estimated rates

• For our discrete samples, the prior odds should just be the ratio of rates
• For things like timeseries, the rate of cleans doesn't really depend on our arbitrary random_rate in the evaluate jobs, so we might want to do something like Rwinsrate/(srate - Rwinstate) as the "rate of samples declared to be glitchy", but then there are issues with target_bounds vs dirty_bounds, etc

We may just want to adopt what is reasonable and correct in the supervised learning context and let the timeseries be whatever they are. As long as we document this well enough (and the rates aren't radically different), the precise choice probably won't matter.