ligo.skymap.distance.conditional_cdf returns NaN for infinite distances

Original report from @andrew.toivonen:

There are a small fraction of incorrect zero/inf distance values being produced

I think that this is referring to an issue surfaced in the sample code for source-frame chirp mass posteriors that I provided in https://git.ligo.org/publications/o4/cbc/task_forces/-/issues/6#note_993182. That code takes an argument bin_edges described as follows:

    bin_edges : numpy.ndarray
        Array of source-frame chirp mass bin edges, **not**
        including the implied leftmost edge at 0 and rightmost
        edge at +infinity.

I excluded chirp masses of 0 and \infty because I was aware of an issue with distance = \infty (corresponding to chirp mass = 0). For that input value, ligo.skymap.distance.conditional_cdf and ligo.skymap.distance.marginal_cdf return NaN instead of the correct finite value.

There are two reasons why these functions return NaN. First, there is a factor in bayestar_distance_conditional_cdf that is of the form r \exp(-r^2), which approaches zero as r \rightarrow \infty, but as written evaluates to inf * 0 which is NaN. Second, the function gsl_sf_log_erfc (also gsl_sf_erfc) returns NaN for infinite or even extremely large arguments.