Binomial: Add benchmark for small n * p

Our previous implementation was very slow (#734). The benchmarks makes
sure we won't have such regressions in the future.

benches/distributions.rs

@@ -203,6 +203,7 @@ distr_float!(distr_gamma_large_shape, f64, Gamma::new(10., 1.0));
 distr_float!(distr_gamma_small_shape, f64, Gamma::new(0.1, 1.0));
 distr_float!(distr_cauchy, f64, Cauchy::new(4.2, 6.9));
 distr_int!(distr_binomial, u64, Binomial::new(20, 0.7));
+distr_int!(distr_binomial_small, u64, Binomial::new(1000000, 1e-30));
 distr_int!(distr_poisson, u64, Poisson::new(4.0));
 distr!(distr_bernoulli, bool, Bernoulli::new(0.18));
 distr_arr!(distr_circle, [f64; 2], UnitCircle::new());