Benchmark algos against publicly available Ouellet puzzles

[jonadsimon]

To determine whether an algo can really pack tightly enough, benchmark against available Ouellet puzzles in wonderword.com

These are 15x15 boards with a very high packing density, ~+20%

(Might already be able to do this for a 15x15? Need to check)

[jonadsimon]

Analyzed Ouellet puzzles within:

• https://docs.google.com/spreadsheets/d/1csESQNB7O_sZgsCGOGhZJpLpZTEvhHjILlNVjmFI_mQ/edit#gid=0
• https://github.com/jonadsimon/wonder-words-generator/blob/main/Analyses/Analyzing%20word%20length%20%26%20letter%20distributions.ipynb

Conclusions:

• Ouellet consistently uses packing of 1.15-1.28x (mostly 1.20-1.25x)
• Denser packings employ larger number of shorter words
• Total # of possible overlaps is biggest determining factor in achievable packing density; "yoga" set was the first/only word set to approach Ouellet's density, and it had a larger # of overlaps than any of his word sets, despite not having the lowest mean word length

[jonadsimon]

Use updated overlappability metric(s) to quantify his word sets on the spreadsheet

[jonadsimon]

Did a lot of analysis, definitely can't handle the packing density of Ouellet's boards, even with similar word lengths and overlap frequencies