Use mutation rate map for missing data, etc

[nspope]

It's nice to be able to use a mutation rate map to account for missing data and/or rate heterogeneity, rather than a global mutation rate. This is straightforward in principle as it only involves remapping edge endpoints to a different coordinate system. However we can end up with zero-span edges (especially for the missing data usecase), in which case the EP updates require special handling. Occasional numerical issues with EP also crop up more frequently.

I've started sorting this out in a feature branch, I'm leaving this issue as reminder to finish it as this is crucial functionality IMO.