Implement CLI parameters for more precise variant likelihoods of MM/INS/DEL

[Krannich479]

Currently, a random variable decides which modification is introduced at a base position with hardcoded likelihoods of MM=0.8, INS=0.1 and DEL=0.1. There should be a better way to control this via the CLI.

[Krannich479]

At the same time, introduce longer runs of INDELs, e.g. using a negative exponential length distribution.

[Krannich479]

Reminder that this should be improved. Currently, the likelihood that a MM/INS/DEL is introduced is coupled to an initial overall likelihood of introducing an error at all. This approach is impractical, slow (unnecessary random floats) and gives a false intuition of the individual mutation likelihoods. The overall likelihood of introducing a mutation should be removed. Also, the option to learn error likelihoods from real data was suggested (thanks to @matthuska )