feature-ode78-ode89

[nickhale]

Implement ode78 and ode89 wrappers (for the new ode solvers in R2021b).

This is the error that will be thrown is a user has an older MATLAB version (which seems sensible)

Unrecognized function or variable 'ode89'.
Error in chebfun/constructODEsol (line 38)
    sol = solver(odefun, tspan, uinit, varargin{:});
Error in chebfun.ode89 (line 29)
[varargout{1:nargout}] = chebfun.constructODEsol(@ode89, varargin{:});
Error in test_ivp (line 34)
[t, y] = chebfun.ode89(@vdp1, [0, 20], [2 ; 0], opts); % CHEBFUN solution 

[aaustin141]

Does anybody object to me merging this? All related tests pass, and I know someone interested in this functionality.

[trefethen]

I'm ok with it if you're confident.

[aaustin141]

The only tests I saw fail were these:

   chebfun2/test_roots.m
   linop/test_eigsRayleigh.m
   misc/test_gpr.m

The failure in chebfun2/test_roots.m occurs on test 25, which already has an extremely loose tolerance. The failure in linop/test_eigsRayleigh.m occurs on test 2. The tolerance would need to be loosened by a factor of about 100 to pass for me, which is a little much, but in an absolute sense, the accuracy is still pretty good (around 1e-9). The failure in misc/test_gpr.m is pretty catastrophic: I only get an error of 1e-4 on test 10, while the tolerance is 1e-10. But it's clear this pull request can't have caused that.