Config option diffuse_wall_bcs_on_init can cause problems with turbulent simulations

[uqngibbo]

I've discovered a weird problem where diffuse_wall_bcs_on_init can put zeros in the flowfield for the Spalart Allmaras field variable nuhat. Zero nuhat is technically an invalid flow state, and although I've tried to make the solver resistant to it, the Preconditioner in particular likes to choke if it's given nuhat=0, and the simulation crashes instantly.

This issue only appeared because of the recent change in 41140df which removes the presteps, which used to smooth out the zeros before they could cause problems.

[uqngibbo]

Could not reproduce.