Add t_start to logZ_noise computation, when t is fixed

The computation of logZ_noise was previously always done at t=0.0 (trigtime) in each detector, instead of the true t_start (e.g. 10M after the peak). This is technically valid under the approximation of stationary noise, but it is more correct to compute logZ_noise on the actual segment used for the analysis.

This MR fixes this, in the case where t and sky-pos do not vary.