Add option to use fallback integration for user material functions

[ChristopherHogan]

Fixes #764.

Allows setting a do_averaging property on a user material function to True (defaults to False) to use fallback_chi1inv_row.

def my_func(p):
  return mp.Medium()

my_func.do_averaging = True
sim = mp.Simulation(default_material=my_func, ...)

This was a little more complicated than the fix described in #764. get_front_object always returns false for a material function, so the if (material_type_equal(mat, mat_behind)) line is never reached.

### TODO
Running the script in issue #764 performs subpixel-averaging when setting do_averaging to True, but fails in the harminv step with

On entry to ZGEBAL parameter number 3 had an illegal value

I traced this back to chi1inv_row getting filled in with nans in geom_epsilon::fallback_chi1inv_row due to a divide by zero here. How should we handle this?