The BCF limit of ~2GB per VCF row causes the parser to return an error.
In this temporary solution we drop the excessive fields, typically these
will be PL and other Number=G tags. in future work the library would
automatically convert such tags into their localized alternatives (LPL).

The removal of large tags by d800f13 inadvertently
changed output for invalid rows, such as the one introduced by 8f782d1.
This was caused by relying on fmt->size==0 never happening - such assumption is
correct, I think, but only for valid tags, not such that failed parsing