Refactor mutation probabilities

[MilesCranmer]

Right now one specifies the mutation probabilities by passing in a long vector of floats. This is super uninterpretable, so this change fixes this. This will also be helpful for adding new mutations, like for the shared-node-utils branch.

With this change, there is now a MutationWeights struct which you can specify as, e.g.,

mutation_weights = MutationWeights(;
    mutate_constant=0.048,
    mutate_operator=0.47,
    add_node=0.79,
    insert_node=5.1,
    delete_node=1.7,
    simplify=0.0020,
    randomize=0.00023,
    do_nothing=0.21,
)

Where these numbers are the relative probability of each mutation (will be normalized afterward). This also makes the code more readable. Now, instead of doing a cumulative sum on the weights manually, and comparing to rand(), a method has been added of type rand(::MutationWeights). This will return a Symbol specifying the mutation chosen. For example:

choice = rand(mutation_weights)
# e.g., :mutate_operator, :add_node

Right now, the parameter is passed in as mutationWeights rather than mutation_weights. I'm planning to refactor all of those in the future to be consistently snake_case at some point, but not now.

@johanbluecreek or @CharFox1 would one of you be willing to take a look?
Thanks!
Miles

[johanbluecreek]

Looks good. It is nice to have a struct for the mutation probabilities instead of having to keep track of the ordering in the vector. The selection of mutation operator is now also much easier to see compared to the cumulative sum approach. I just left some comments.

[MilesCranmer]

I feel like sample(mutations; weights=options.mutation_weightings) might be more intuitive than rand(mutation_weightings).

[MilesCranmer]

Thanks for the review!