PhenomD and PhenomPv2 discontinuity

[zipengwang98]

several things I noticed while working on IMRphenomPv2:

• At the 1e-5~1e-20 matching level, a constant time shift between the waveform matters. in PhenomP they calculate the time shift using 3rd order spline approximation of the waveform around fRD. By reproducing their routine (instead of taking derivatives using autodiff) I increased the matching between ripple and lalsuite from 1e-5 to 1e-10.
• There is a discrepancy between phenomD and phenomPv2 in lalsuite. This can be seen by matching lal's phenomD and lal's phenomP with no in-plane spins: the matching is bad.
• There is also a discontinuity in lal's phenomPv2 right at in-plane spin = 0 (not in ripple as far as I know). The waveform with s_x = 0.0001 is different (a constant phase shift maybe?) from that with s_x = 0.