Feature: Parallel iterations

[MashAliK]

Add parallel iterations along with a couple of bug fixes/improvements. I would consider this a pretty important feature because of the significant speedup it provides to training. This is the implementation that I tried and has worked for my use cases.

Primary changes:

• Use concurrent.futures in controller.py to spawn processes for performing the iterations of training
• Moved most of the iteration logic inside the function run_iteration_sync in a new file iteration.py which is run by the workers to allow concurrent execution
• Pass config into spawned processes to initialize new instances of LLMEnsemble and ProgramDatabase (these classes are not pickleable so this is the best approach I could think of to use these classes in each of the worker processes)
• As a consequence of this approach a snapshot of the database needs to be saved in the root process every iteration and each new process needs to load it

Minor changes:

• The calculate_edit_distance function was crashing the database when I was using it and since there's already libraries that do this routine I ended up using one of them (levenshtein)
• Replaced the use of the Levenshtein distance with a ratio in _calculate_island_diversity since it's easier to read
• Replaced the use of the Levenshtein distance with a ratio in _calculate_feature_coords since it's normalized for code length
• Introduced allowed_population_overflow since otherwise the database was adding and removing a program every iteration when it reach the allowed program limit
• Added logging for MAP-Elite features

[MashAliK]

Solves: #32
I believe it also contains the solution to #60 because I remember moving the call to _enforce_population_limit later in add fixed an error where the program was removed from the list while it was still being added
Taking a closer look this bug looks different then the one I saw (I encountered mine when calling add, not sample)

[codelion]

Thanks for contributing can you rebase from main, I can test and then review this PR.