Ascertainment correction & missing data

[rbouckaert]

The ascertainment correction when there is missing data is not quite right: we condition on all sites being constant, but in fact for those sites that made it into the alignment but have missing data, we really should condition on all sites that have data being constant: it is very well possible that the missing data varies from the rest of the data for a site.

Unfortunately, calculating such a correction means adding patterns for all combinations of missing data, making runs potentially a lot slower.