Use AMPLFI posteriors to compute p_astro over population model range #466

The per-class signal rate for class $c$ (BBH, BNS, NSBH) is

$$ R_c = \int p(d|\theta) \pi_c(\theta) d\theta $$

where $\pi_c$ is the class sub-population from #465. Rather than approximating this with a fixed mass fraction, estimate it via importance sampling over AMPLFI's posterior samples $\theta_i \sim q(\theta|d)$:

$$ R_c \cdot \frac{1}{N} \sum_i \frac{\pi_c(\theta_i)}{q(\theta_i)} $$

The Monte Carlo variance of the importance weights gives error bars. Implement in libs/p_astro/ and wire into online/subprocesses/p_astro.py.