Return negative loglikelihood function.
Notes
This is -sum(log pdf(x, theta), axis=0) where theta are the parameters (including loc and scale).
scipy.stats.rv_continuous.fit_loc_scale
scipy.stats.rv_discrete