TODO

For the root treatments with stationary distributions for functions
that modify parameter vectors (currently only classe) be careful that
the stationary distribution is correct.  This is not the case right
now for classe.

ROOT.EQUI check is not being done for t/td/split models as the
rootfunc hides it.  Need to address this.

bd.split is really ugly.

The entire history bit (as with most of the plotting) is a bit of a
gong show.

Non ultrametric tree support should be included.  If the info list
includes a "non-ultrametric-ok" element (or just look at idx.e) we
could generate a function that only looks at 'len', not at t0, so that
rather than doing
  branches <- function(y, len, pars, t0, idx)
    ode(y, c(t0, t0+len), pars)
we do 
  branches <- function(y, len, pars, t0, idx)
    ode(y, c(0, len), pars)
This would break time-dependent mk2 though.