Here's my solution!
This was surprisingly not too difficult - the key was not to think about the number of permutations or the number of letters but to think about how the letters match to what is in your hand, do a few filtering joins, and then think about the scoring only as a function of the tiles you have, not the value of the actual words created. Granted, my solution is not the most efficient thing in the world because I tokenize the subset of qualifying words twice where if I spent more time on it, I could probably eliminate this step.
But this runs fairly quickly (all three tasks run in under 30 seconds) and its configured that if you have super duper special Scrabble with even more than 2 wildcard tiles, you can change two tools to incorporate even more wildcards.
