Check Jacobian gradient before correcting higher derivatives

[simonsticko]

Correcting the 2nd and 3rd derivatives of the shape functions with the derivatives of the Jacobian is expensive. #9530 adds a function
higher_derivatives_need_correcting in fe/fe_poly.templates.h
that returns whether the correction terms need to be computed. Currently this function only checks if the mapping is Cartesian, but a more general condition would be to check the value of the jacobian_grads_pushed_forward at the quadrature points. Precisely what condition should be checked is open for discussion.

[bangerth]

The condition to check is whether the norm of jacobian_grads_pushed_forward is << than the norm of jacobian divided by some length scale (e.g., the diameter of the cell we're on).

[masterleinad]

I don't think this is crucial for the release and would move the milestone unless there are volunteers to tackle it.

[peterrum]

@simonsticko How do we want to continue here?

[kronbichler]

I guess this is not critical, so we can postpone the target milestone (and if we happen to have time for this one we can still work on it).

[kronbichler]

I think we have substantially reduced costs in MappingQ related functions by #15035 and #15125 and I consider this issue addressed. Of course it would be nice to get further progress also for other mappings, but I suggest to remove it from the current target milestone, given how long this has been around unaddressed.