Added 2nd-order Heun's method and midpoint method for ODE sampling

[FENRlR]

The code

dphi_dt = self.estimator(x, mask, mu, t, cond, training=training)

        if guidance_scale > 0.0:
            mu_avg = mu.mean(2, keepdims=True).expand_as(mu)
            dphi_avg = self.estimator(x, mask, mu_avg, t, cond, training=training)
            dphi_dt = dphi_dt + guidance_scale * (dphi_dt - dphi_avg)

was separated to def func_dphi_dt for ease of reuse for both methods.

[p0p4k]

Thank you!