Resolve "Plot_time_domain_data and time of coalescence for waveform SEOBNRv4"

This MR fixes the correction of the epoch in time domain waveforms when using a segment duration that is not a power of two, e.g., SEOBNRv4 with 5s duration.

I don't think that this will impact O3 analyses, but I wanted to check. Bilby runs are just frequency domain, right? @christopher-berry @cjhaster @gregory.ashton

Closes #551 (closed)

bilby/gw/source.py

-from __future__ import division, print_function
-
 import numpy as np
 
 from ..core import utils
@@ -408,11 +406,10 @@ def _base_lal_cbc_fd_waveform(
     h_cross *= frequency_bounds
 
     if wf_func == lalsim_SimInspiralFD:
-        dt = 1. / delta_frequency + (hplus.epoch.gpsSeconds + hplus.epoch.gpsNanoSeconds * 1e-9)
-        h_plus *= np.exp(
-            -1j * 2 * np.pi * dt * frequency_array)
-        h_cross *= np.exp(
-            -1j * 2 * np.pi * dt * frequency_array)
+        dt = 1 / hplus.deltaF + (hplus.epoch.gpsSeconds + hplus.epoch.gpsNanoSeconds * 1e-9)
+        time_shift = np.exp(-1j * 2 * np.pi * dt * frequency_array[frequency_bounds])
+        h_plus[frequency_bounds] *= time_shift
+        h_cross[frequency_bounds] *= time_shift
 
     return dict(plus=h_plus, cross=h_cross)