As part of my efforts to investigate space leaks, I ended up at the fold_lookahead function, which I rewrote as below. I offer this as an entirely separate patch because there are good reasons you might want to not merge it. Advantages:

• Reduces stack usage from > 1Mb to < 1Kb.
• Algorithm more directly states what it does.
• Better time complexity - previously it was a combination of insertion sort on the Int and a weird order-changing nub construction on the Lr0Item.
• Passes all the tests.

Disadvantages:

• Has no meaningful time difference - my tests are with 0.03s.
• Produces different output since the list is in a different order. I believe the results are equivalent, but you will need to say for sure.