Exclude gaps from chisq calculation

To calculate chisq, we try different offsets in the ifo's data. If we hit a gap, it technically doesn't contribute any data, but the data point is still effectively included in the divisor: autocorr_norm.

An attempt to exclude these data points from the divisor was made, but it didn't make any significant impact.
See spiir-production-Oct27-dev and the corresponding changes in tdavies__remove_snr_constraint (as of f39069d3)

This will require some investigation. We could go with the approach above, but would need to account for the calculation of autocorr_norm (the version above assumed it equalled the number of data points, which it does not). But we could try alternatives. For example, maybe we could select different points each time we hit a gap.