automaton.hpp: Count kmers even if they are not new
This seem weird and is a bit hacky. But the purpose is not to disadvantage a real germline (eg. IGH) compared to a fake germline (unexpected). As there are more ambiguous k-mers in the unexpected germline there will be less IGH k-mers hence a smaller index load for IGH in the unexpected germline rather than the real one. See also #2651.