Marginalize over supernova M parameter

[joezuntz]

This adds analytic marginalization over the supernova absolute magnitude M in the pantheon likelihood.

[joezuntz]

Comparison plots for pantheon example (without Riess or H0) with and without marginalization

[tamarastro]

Based on the equations in Goliath et al. 2001 the constant term on line 37 and line 148 is flipped, and has a superfluous 0.5. Although on line 156 you multiply D by 2 so that gets rid of the superfluous 0.5 (but I think you could drop both the 0.5 and the 2). Then on line 156 the sign of -2 * D should be +2 * D I think.
Line 37:
self.marginalize_const_term = 0.5 * np.log(2 * np.pi / self.marginalize_c_term)
->
self.marginalize_const_term = np.log( self.marginalize_c_term / ( 2 * np.pi ) )
Line 148:
D = (1/2) log(2 pi / C)
->
D = log(C / 2 pi)
Line 156:
change = -B ** 2 / C - 2 * D
->
change = -B ** 2 / C + D