'bzr pull' could use update_basis_by_delta instead of set_parent_trees

Currently, bzr pull does an iter_changes() style to merge the new contents into the working tree, but then calls "tree.set_parent_trees()" which is an O(tree) operation.

In a case where there aren't any merges, we should be able to use tree.update_basis_by_delta(). Also, the future plans call for not caching merged parents in the dirstate (we don't use it to speed up commit, which was the primary use case). At that point, we could always use a delta to update our metadata.

This currently (bzr-2.4 or bzr-2.3.2) costs us about 15s on a 70k entry dirstate, because of the time it takes to load all the pages of the CHKInventory.