sample command silently produces incorrect empty reads #213

Hi, I have a fastq with some empty reads like so:

@name

+

@other
AAA
+
999

Running seqtk sample bad.fastq 0.9 on this produces:

>name

@other
AAA
+
999

which is no longer valid fastq, as the @name read has changed format. This breaks lots of downstream tools. Is it possible to keep it in the original format? I'd like to preserve these if sampled because there might be a read pair which is nonempty.