[Alignments] Ill defined reversed offset upon in/dels

[qmarcou]

Insertions or deletions in sequences introduce a shift in the final offset which is counter intuitive when the offset is defined as a reversed offset.
The ad-hoc fix is to widen a bit the constraint on offsets to compensate for a few in/dels, however this is silly and make the aligner lose some precision.