timeasc/timedesc node orderings only hold *within* a particular root

[jeromekelleher]

If we have two disconnected subtrees in a tree, the tree.nodes(order="timeasc") function currently visits nodes in one subtree first in time order, then the other.

It would be more consistent and logical if all nodes in the tree were visited in time order.

[jeromekelleher]

Also we're currently defining it like this:

self.nodes(u, order="levelorder"), key=lambda u: (self.time(u), u) 

which means we sort by time first then by node ID. I would have thought it simpler/more logical to leave the natural preorder tree ordering within a node. (The "levelorder" isn't needed either, since we're totally ordering)

[jeromekelleher]

Also true for levelorder.

[benjeffery]

All the sorting methods go root-by-root, I assume for the others this is OK as they are topological.

[jeromekelleher]

I'm working through traversal semantics right now as part of #1704 - I think it'll be easiest to change it to be global rather than per root. This will be a good time to change it, since we'll be doing the virtual root as part of 0.4.0