WT.iter_changes(arbitrary_tree) should use chaining

[John A Meinel]

Right now, bzr has 2 fast iter_changes paths. RevTree.iter_changes(RevTree) can use the CHK logic to avoid walking the whole tree to compute the difference. And WT.iter_changes(tree-in-dirstate) can walk line-by-line in the dirstate to be very fast as well. (And in general, we have to walk the real-working-tree to see if there are any recent changes anyway.)
The dirstate layout is not similar to how the data is stored in the repository (dirstate is ordered by path, repo is ordered ~by file-id.).
However, we should be able to composite 2 iter_changes output together. So doing "WT.iter_changes(arbitrary_tree)" can do:
 WT.iter_changes(WT.basis_tree()) + WT.basis_tree().iter_changes(arbitrary_tree)

[John A Meinel (jameinel)]

This would be relevant for "bzr revert/status/diff -r -10". I'm not sure if that is common enough that we should focus on it, though. Implementing chained iter_changes() is a fair amount of logic, and needs very careful edge-case handling.