Deep Combinators

We have several different "deep" combinators that can dig into ASTs.

Issues

With unfold, there is no clear winner.

• unfold matches as many arguments as possible, then tries to normalize them.

• This mirrors what a human does with unfold (aka fold/unfold methodology)

• any-bu matches the depth 0 before depth 1,2.., so does not get the best match

• any-tu matches, but then descend into unfolded function (perhaps this is okay?)

• prune-td matches the best "unfold match", but then does not handle the arguments to the function being unfolded. So

  f . (g . h)

Only the outermost . gets unfolded.