bilby/gw/likelihood.py: speed up the spline interpolation in ROQ

This MR speeds up the spline interpolation of d_inner_h_tc_array in ROQ.

Since it is apriori known that the time interval of samples is constant and 5 samples are used, the spline method tuned for that case is much faster than scipy.interpolate.interp1d. The algorithm is described in https://dcc.ligo.org/LIGO-T2100224. This change speeds up the interpolation by a factor of ~8. This can be a significant speedup for compact ROQ bases. For example, this change speeds up the overall ROQ likelihood evaluation by a factor of ~2 for PhenomD bases with size of ~70.