Tracing fixes and fast fluxsurface tracing

Several fixes for tracing and implement Equilibrium.trace_flux_surface which uses optimized normal to grad psi. This flux surface tracing offers much finer resolution in comparison to the find_contour approach with several times lower runtime and memory requirements (comparing with a similar resolution in the contour approach). As a bonus it calculates FluxSurface.cum_length as a byproduct.

You can play around with it and perhaps use it as the default flux surface generation if you find it useful.

I base this PR on the sumtraic2019 branch to prevent a large PR, but that can be changed.