Improve the verse segmentation algorithm by setting a threshold for quality using alignment scores, and then if the threshold is not met, combining them into a single verse range, rather than trying to split them. We could also use length as an indicator (so a verse range doesnβt get too long for NLLB).