Factor out common functions in notebooks + Perf Improvements + max-splits heuristic

[vtjeng]

• We factor out common functions in performance.ipynb and wordle_solver.ipynb into a single utils file. This will make code reviews in the future simpler, since diffs of .jl files display better than those for .ipynb files.
• Significantly improve the performance of the code in performance.ipynb by using a 2d-array to store the precomputed scores rather than a dictionary, and by making the score a UInt8, which takes up less space.
  • w/o hard mode, computing using all words goes from 900s -> 200s
  • Implementation:
    • get_group_sizes, find_move, trim_pool, apply_strategy and get_num_turns make use of type polymorphism to handle either an Int (corresponding to the index of the word in the list of all words) or a String (the word itself).
    • In addition, to get the score to fit in a UInt8, we treat the scores as ternary (e.g. 22222 in base 3 is 242 in base 10).
• Add charts for max-splits heuristic + hard mode with all guesses.
  • Max-splits seems to perform consistently best on every metric for hard mode.

NOTE: You'll see that there's some differences in our histograms. I haven't been able to dig in to exactly what the issues are, but I suspect it has to do with the change to tiebreaks (from lexical order on strings to the position of the words in the guessing list).