Code quality improvements for `fuzzing/valuegeneration`

- It's a bit clunky to have RandomValueGenerator implement ValueMutator, since that's not its purpose. Right now all its Mutate functions are dummy functions that return the input. Do we need to implement this at all; can we just remove these methods?
- It's a bit clunky to have MutationalValueGenerator implement ValueGenerator, since that's not its purpose. Right now its Generate methods call mutate on a blank value, basically just repeating the work of RandomValueGenerator. Do we need to implement this at all? Can we just remove these methods and use RandomValueGenerator instead?
- EncodeJSONArgumentsToMap, EncodeJSONArgumentsToSlice, EncodeABIArgumentsToString, and so on should probably be somewhere else
- MutationalValueGenerator.MutateArray does nothing and has a big TODO in it: `TODO: Apply array structure mutations (swap, insert, delete)`
- I think the parentheses [here](https://github.com/crytic/medusa/blob/master/fuzzing/valuegeneration/generator_mutational.go#L109C2-L109C134) are wrong: `mutationCount := g.randomProvider.Intn(((g.config.MaxMutationRounds - g.config.MinMutationRounds) + 1) + g.config.MinMutationRounds)`

Start in fuzzing/valuegeneration, especially generator_mutational.go and the RandomValueGenerator and MutationalValueGenerator implementations. Trace the generator and mutator interfaces, the argument-encoding helpers, and the mutation-count calculation before deciding which responsibilities can be removed or relocated. Done means the responsibilities are separated cleanly, the mutation-count expression is corrected, and the array-mutation TODO is either implemented or explicitly resolved.

Written by the indexing model from the issue text.