Pre-compute loss function in finite_horizon

[LarrySnyder]

Tried working on this, but it didn't seem to speed things up much. Why?

Added this code right before main loop:

	# Pre-calculate standard normal loss function values.
	min_z, max_z, step_z = -4, 4, 0.01
	loss_table = lf.standard_normal_loss_dict(start=min_z, stop=max_z, step=step_z)
	comp_table = lf.standard_normal_loss_dict(start=min_z, stop=max_z, step=step_z, complementary=True)

And this code when n(y) and \bar{n}(y) are calculated:

				if [...some option is set...]:
					z = (y - demand_mean[t]) / demand_sd[t]
					if z < min_z:
						n = demand_mean[t] - y
						n_bar = 0
					elif z > max_z:
						n = 0
						n_bar = y - demand_mean[t]
					else:
						n = nearest_dict_value(z, loss_table) * demand_sd[t]
						n_bar = nearest_dict_value(z, comp_table) * demand_sd[t]
				else:
					n, n_bar = lf.normal_loss(y, demand_mean[t], demand_sd[t])

Maybe nearest_dict_value is too slow?