xyst

Add better src to ZalCG

Better than only adding lumped-mass src at the end of the TG time step,
yielding lower errors, but still unstable with vort 48M. Could be due to
non-divergence-free numerics of the first TG half step.