All that is needed is to align the sequences within a formal aligner like MAFFT or muscle. Then pass that through the PDGSpider converter. Arlequin should just take it from there.
Manual annotation of the alignment is normal e.g.
seqoTron and is traditionally regarded as the best approach rather than do this algorithmically. Thus I'm suggesting (strongly suggesting) do this prior the PGDSpider conversion. If an alignment has not been performed Arlequin should throw an error.
If there are trailing gaps (indels or '-') at the beginning and end of the alignment Arlequin will/should ignore these. You can of course simply cut them out with the alignment editor (Mega X will also work for alignment editing). The important point is that Arlequin must correctly identify homologues its an allele frequency package and nothing more. If there are errors in the alignment Arlequin will believe there is population structure when it could be absent.
I can't honestly believe that Arlequin will count an indel '-' as an informative site. It should just ignore those sites, because it is a very respectable package that has been around for years and years.
Okay, I am now signing off for Christmas... with occasional exceptions.