more adaptive tolerance on gradient squared norm

[liminchen]

Although we normalized the distortion energy Ed, the gradient of this energy should still be relevant to the surface area.
It would be meaningful to see whether an adaptive tolerance on the gradient norm like the one in BCQN would further improve the adaptivity of the stopping criteria of geometry steps.