Empirical mutation probability

I have a suspicion that empirical mutation probabilities are different from the theoretical. What I mean:
[Here ](https://github.com/aimclub/GOLEM/blob/main/golem/core/optimisers/genetic/operators/mutation.py#L102) when we call mutation, we iterative choose mutation until it can be applied. Obviously that mutation, f.e. "parameter_change" always can be applied because it can't performs changes affect on verification. But mutations "add_parent", "add_intermediate_node" can make pipeline incorrect. Due to above facts we can stuck in situation, when almost all mutations are "parameters change" (critical in Fedot ts forecasting task when there are a lot of constraints put on a pipeline structure).

What I assume to do:

- [ ] make experiments in GOLEM and Fedot particularly (in different tasks). Find is there a real problem
- [ ] change logic of mutation, for example give structural mutations more tries to apply it's mutations