Plot_time_domain_data and time of coalescence for waveform SEOBNRv4

I came across the following unexpected behaviour when plotting time domain data with bilby. When plotting GW strain with for the waveform model 'SEOBNRv4', for a duration of 5s, the chirp signal is not centered around 0 anymore. This can be seen for example when modifying the duration to 5s, setting tilt_1 and tilt_2 to zero, as well as putting the waveform approximant to 'SEOBNRv4' in https://git.ligo.org/lscsoft/bilby/-/blob/master/examples/gw_examples/injection_examples/plot_time_domain_data.py When widening the values for which the time domain is plotted one can see that the chirp signal is shifted to higher or lower values (depending on the exact value of the duration).

In principle this is a problem, since this can affect the computed SNR (since parts of the signal are cut away). This problem is not present for waveforms like 'IMRPhenomPv2'.

If I understood correctly, a similar problem was addressed in #445 (closed) I took the script from #445 (comment 138188) and plotted the hplus_INSPIRAL_FD.epoch values for different waveform models vs the duration (denoted as T in the script).

It appears to me that all durations which are powers of 2 are special values for the SEOBNR* models (and this might also lead to the signal being not injected at 0 for the 'SEOBNR*' family).

I hope that I have not misunderstood how the function plot_time_domain_data is supposed to work. In case that this is a wanted feature, please forgive my ignorance.