Refine velocity with sub-pixel CCF fitting for the closest grid point

[andycasey]

The initial CCF velocities are systematically lower than GUESS by a few km/s, likely due to the fact that the initial CCF routine only looks for the pixel of the strongest peak, without actually fitting the CCF profile. The routine should identify the best-fitting spectrum from the grid and perform the cross-correlation again, this time with sub-pixel fitting.