bilby/gw/detector/calibration.py: extrapolate calibration errors for smooth transition

This fixes #637 (closed). Since this is a corner case, I am happy to close this if extrapolation is not preferred for other reasons. This also benefits the multiband likelihood, as it requests waveform values beyond the high-frequency cutoff and assumes smoothness of waveform there, although log-likelihood errors introduced by the abrupt termination of calibration errors are typically below unity.

added <10 lines label

@soichiro.morisaki sorry for leaving this so long. Is this important for online PE?

I don't think this is critical one for online PE. There could be some BNS events for which we find glitch at low frequency and we want to use f_low higher than the cutoff of ROQ for manual follow ups. Smooth transition of calibration uncertainties will be preferred in that case.

mentioned in merge request !1241 (merged)

This is included in !1241 (merged), so I am closing this.

closed