Allow to remove features before starting the first population.

[jonlachmann]

When starting gmjmcmc it should be possible to:

• Manually specify the set of covariates that should be included in the first population.
• Randomly select a set number of features for the first population.

This allows the algorithm to work with "wide" data where it is not feasible to include all covariates in the initial population.

[jonlachmann]

https://github.com/jonlachmann/GMJMCMC/commit/949a2d748e9f9469a8736960280be175df3a973f