Add proper adaptive timestepping to elastic simulations

[ilateur]

Problem

Adaptive time stepping did not properly work with magnetoelastic simulations, because the absolute error of all Variables was used to gauge accuracy. An absolute error of 1e-5 or lower was seen as good enough for displacements of the order of ~1e-14 m and velocities of ~1e-4 m/s.

Solution

The essential difference is moving the maxError of adaptive time stepping from TimeSolver to DynamicEquation, so each equation has its own error to adhere to. This is not user-accessible, so the 3 relevant max errors can be set via the (Ferro)Magnet, with automatic telepathic pointer-communication (I'm not the biggest fan of this solution). This is the only (or maybe not) solver setting outside of TimeSolver :( but it also has a physical unit and is physics dependent.

As a consequence, the Relaxer cannot directly set maxError of the TimeSolver anymore. This is not a problem, because it needs to create new DynamicEquations anyway, so it can also assign a pointer to its own maxerr variable, then scale that.

Let me know if you have a better solution though.

Potential To-Do's

• take a look at sensibleTimeStep()
• may need to tweak default values 1e-18 and 1e-7 of displacementMaxError and velocityMaxError.
• make starting value of maxerr_ in Relaxer user-assignable? This used to start at maxError of TimeSolver.
• clean up Relaxer?
• clean up raw pointer logic?

[DiegoDeGusem]

The magnetization max error is a ferromagnet property, so it is possible to set 2 different errors in an antiferromagnet? That's a bit strange 🤔
Maybe this could also benefit from some tests and/or an example?
I don't immediately see a better way of handling this, so this method is fine by me I think

[ilateur]

Do you have a suggestion for any test?
As examples I figured it was fine to remove the fixed time step from all (magneto)elastic examples.

[ilateur]

The implementation has been done in a different way, see pull request #60.