refactor(lib): Allow fitting whole population at once

The fitness function gets fit_population function. The default
implementation is going over fit() one by one, but it can be
reimplemented by specific types.

feat(lib): constraints along with evolutionary strategies

chore: fix warnings

feat(lib): allow modifying fitness function in evolution_algorithm

chore: move population structs to separate module

feat: split params out of PerturbationOperator to allow for non-'static PerturbationOperator

This means PerturbationOperators can now hold non-'static references.
Right away, utilize this in the Random2OptPerturbation for the TSPInstance.

feat: add a lot of algorithm showcases

feat(lib): add evaluation count to EA stats

fix(lib): do not take mut self in population.iter()

feat(lib): add roulette wheel selection

feat(lib): add random search

feat(lib): add n consecutive bit flip perturbation for binary string

feat(lib): add two point and n point crossovers

refactor(lib): implement generic evolutionary strategy using .wrapped_mut

Thanks to wrapped_mut now the one_to_five strategy can be implemented
without relying on BoundedPerturbation etc.

feat: add possibility to evolve strategy during evolution_algorithm

refactor: generalize apply_to_mutations

refactor(lib): allow modifying mutation perturbation

refactor(lib): add methods to get wrapped perturbations instead of traits.

Remove WrapperPerturbation and ListWrapperPerturbation in favor of
wrapped and wrapped_mut functions on PerturbationOperator. This
allows for easier handling as it's not easy to downcast Any into
WrapperPerturbation or ListWrapperPerturbation. So instead
always return dyn PerturbationOperator instead of dyn Any.
Use dyn Any only for downstasting to concrete types.

This finally allows for apply_to_mutations to be implemented
under MutationPerturbation, allowing to modify the arguments.

chore: do not clone for no reason in local search

feat: add into_iter for population