xml parsing: start traversing properties too

This requires us to change the way we match things up, because now we
can have things like

   dives.dive.sample.event.time

and

   dives.dive.sample.time

and they are different things (that "sample.event.time" is a 'time'
property of the 'event').

Now, this is always going to be ambiguous, since our linearized name of
the xml doesn't really care whether it's a xml node "child" or a
"property", but quite frankly, I don't care. XML just isn't worth the pain.

In fact, maybe this ambiguity can end up being a good thing.  We will
parse these two different lines of XML the same way:

  <dive><sample><time>50</time><depth>10.8</depth></sample></dive>

  <dive><sample time="50" depth="10.8"></sample></dive>

and the attribute approach seems to be the nicer one.  Maybe I'll use
that for the output format.

Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>