Sort perf improvements

[dpbutter]

Per our discussion last year, here is a better version of sort_sum that offloads the sorting into a new subroutine in tree.hh.

[kpeeters]

What was the motivation to introduce a new sort_subtree into tree.hh, when there is already sort?

[kpeeters]

For reference: the discussion referred to above is in PR #286.

[dpbutter]

The existing sort seems dependent on the compare_nodes class, which only compares on the data of the node in question (i.e. the root node of the subtree to be sorted). So without breaking the tree.hh API, it seemed I couldn't use the existing sort to accomplish this.