`unfoldTreeM_BF` and `unfoldForestM_BF` previously used a
queue-based algorithm from a note by Okasaki. That same
note indicates that the level-wise approach tends to be slightly
faster, and my own informal tests suggest the same. Both solutions
suffer from a potential space leak, retaining the seed lists
longer than necessary in order to later use their lengths. It
remains to be seen whether this leak can be plugged without
a speed penalty.