vg deconstruct AT fields not necessarily consistent with different levels

[glennhickey]

Each site is converted independently. In the event of multiple traversals through a site for a given haplotype it just chooses one (I think) and uses it. If there are multiple traversals with different alleles it will flag a "conflict" but otherwise, if the traversals spell the same alleles it just carries on.

But... when considering traversals in the AT fields, this can lead to trouble. If there are 2 equivalent top-level traversals, it chooses 1, but there's nothing to force it to choose consistent traversals for the child sites. This can lead to situations where a child traversal's AT path is not a subset of its parent.

Hat tip to @wwliao for flagging this.