Permit fastq output to create empty FASTQ records for seq "*".

[jkbonfield]

This is rather questionable, but htslib can read in empty fastq records and generates "*" for SEQ and QUAL. The same is true vice versa now. Eg:

name    4   *   0   0   *   *   0   0   *   *

becomes

@name

+

Bwa mem and minimap2 can both read these fastq entries too, although the SAM output is bugged as it outputs an empty field instead of "*" for SEQ. (Note this is still readable by htslib and interpreted the same).

Potential reasons for accepting this:

• When dealing with paired data, we don't want to output a differing number of records from samtools fastq if read1 has seq and read2 has "*".

  Note as this filtered at the htslib layer, it's not considered as a singleton so fastq -s won't rescue this.

• At least some aligners apparently support this format. Although inevitably they just produce unmapped data.

• Arguably this is a case of silly input => silly output!

• Users can manually elect to "samtools view -e 'length(seq) > 0'" before using samtools fastq, which then fixes the not-a-singleton problem.

• It converts samtools fastq output back to how it was in pre 1.13 era, where we rewrote it to use htslib's interfaces.

Potential reason to reject:

• It may yield output which trips up some poorly written tools.

Fixes samtools/samtools#1799

[daviesrob]

On discussion, we decided that avoiding mis-paired reads was preferable to avoiding empty sequences. I suspect this rarely happens, but I guess we may find out if anyone notices...