Collisions in the adapter index should be handled more gracefully

[marcelm]

Given these two prefix adapters (plus some more so that the indexing logic is triggered) and allowing two errors:

^AAGCGCCAT
^AGGCATCAT

Cutadapt does not trim this query sequence at all:

ANGCATCATAAAAAAAAAA

And for this query, the first adapter is trimmed although the second one would have fewer errors:

AAGCATCATAAAAAAAAAA

Here are the alignments for the two adapters against the first query:

AAGCGCCAT
 X  XX
ANGCATCATAAAAAAAAAA

AGGCATCAT
 X       
ANGCATCATAAAAAAAAAA

There is a warning when creating the index that there are exist two sequences that cannot be matched uniquely. However, for the above sequences, it is clear which one is the better match.

This can lead to incorrect demultiplexing for adapters that can collide (even if the sequence that occurs in the read can unambiguously be assigned).