Aitken-Neville uses wrong controller

[termi-official]

Describe the example

When refactoring the controller interface I noticed that Aitken-Neville uses a PI controller instead of a custom extrapolation controller. IIUC this is not correct, as changing the coefficients in this controller class permanently yields sub-optimal control performance. It seems that it was missed in one of the previous updates of the sublibrary, as the error estimation part in perform step looks differently that the implementation of all other extrapolation algorithms (c.f.

OrdinaryDiffEq.jl/lib/OrdinaryDiffEqExtrapolation/src/extrapolation_perform_step.jl

Lines 107 to 110 in 10501ef

	dtpropose = step_accept_controller!(
	integrator, alg,
	stepsize_controller!(integrator, alg)
	)

).