Using powers of two also matches what we do in almost all cases in an
actual evolution.

The detailed roundoff behavior shouldn't matter in the regimes we care
about, but the symmetric form is convenient for tests where negating
all the inputs now exactly negates the output.

This does not change the StepChoosers to use new features of the
interface, although the new slab-size goal can come into effect with
the LimitTimeStep control system action (as desired).

Instead of making people figure out which to pass for backwards
evolutions, just accept either.

* Do nothing when not active
* Don't set limit as a new goal

StepToTimes no longer has roundoff problems.

Much more common operation than an unconditional size increase.

Like Constant, but will not increase the step size if it is not being
changed for other reasons.

Most appearances of Constant in the examples were not useful, so
removed instead of kept or replaced.

Always clearly indicate whether the class provides a goal or a limit.