Propagator tree compilation to make sure the statistics are correct

Currently e.g. given a propagator with crossover followed by single point mutation, each with a set probability:
First crossover is applied with its probability and then mutation with an independent probability.
If in the end no propagator was applied, then a completely new individual is randomly generated.
This skews the statistics and might not be what the user expects.

We should either organize the propagator differently to behave closer to expectation and/or communicate the resulting statistics more clearly.

No files or tests are named. Start by tracing how the propagator tree applies crossover and mutation probabilities and when it falls back to generating a new individual. Clarify the expected behavior and statistics first; the work is done when the chosen behavior is implemented and the resulting statistics are clearly communicated.

Written by the indexing model from the issue text.