Fix some issues with distance and phase marginalization

The changes in !858 (merged) require a corresponding set of modifications to the joint distance+phase marginalized likelihood. This fixes the issue that @moritz.huebner ran into when running the GW150914.py example.