Consider initializing spin components for restricted-spin approximants
Out of all the spin d.o.f., aligned-spin approximants only produce samples for
spin_2z. Zero-spin approximants don't output samples for any spin parameters. PESummary currently handles this by producing arrays that are zero for all unmodeled spin values. However, this is not accurate for all parameters.
In aligned-spin approximants:
a_1 = abs(spin_1z);
a_2 = abs(spin_2z);
cos_tilt_1 = sign(spin_1z);
cos_tilt_2 = sign(spin_2z);
spin_1x = spin_1y = spin_2x = spin_2y = 0.
In zero-spin approximants, everything is right to be zero except for
cos_tilt_2, which can take any values, but something like
np.nan is probably more accurate.