DEV: neaten calibration marginalization to enable reconstruction

As an alternative to lscsoft/bilby!1234 (diffs) I removed the obstacle to the reconstruction in the way the likelihood is calculated. It turns out this was as simple as changing

return logsumexp(log_l_tc_array, b=time_prior_array)

to

return logsumexp(log_l_tc_array, b=time_prior_array, axis=-1)

The other changes are condensing two methods into one that allowed some of the control flow to be made more consistent.