reimplement stellar boundary conditions

[zingale]

Sometimes we only model a portion of a star, with the surface outside of the computational domain. We used to have the ability to spherically average the star on the grid and use this profile to initialize the ghost cells. This was removed in 5e8a228fc4e3b05fec53acca2d2a7dd2de0ab290

We should add this back in. Perhaps extending the model via HSE to get the corners of the domain correct too.