Is Gcapp.asymdif asymptotically suboptimal? More broadly, should we reconsider the IR for standards?

[nlevenson]

Looks like it's O(a * b), where a and b are the sets of nodes in each tree.

Also, it's always worst case because we have no mechanism to terminate DFS early.

Improvable? Makes you wonder if the intermediate representation for standards that Ground Control uses under the hood should be a BST sorted by the hash value of each node? This would add some constant time factor at construction time, but would let us search over standards in O(h)...