bugfix: HD line should be printed before SQ

[nh13]

Fixes a long standing bug, but made more apparent in https://github.com/lh3/bwa/pull/336. Thank @jmarshall for pointing this out.

I tested this by on my local machine three ways.

1. Not specifying -H, so the default header line should be emitted at the start of the SAM:

   ./bwa mem phix.fasta phix.fastq 
   [M::bwa_idx_load_from_disk] read 0 ALT contigs
   @HD VN:1.5  SO:unsorted GO:query
   @SQ SN:gi|9626372|ref|NC_001422.1|  LN:5386
   ...

2. Specifying -H with @HD, so the user provided header line should be emitted at the start of the SAM:

   ./bwa mem -H '@HD  VN:1.4  SO:unsorted     GO:query'  phix.fasta phix.fastq 
   [M::bwa_idx_load_from_disk] read 0 ALT contigs
   @HD VN:1.4  SO:unsorted GO:query
   @SQ SN:gi|9626372|ref|NC_001422.1|  LN:5386
   ...

3. Specifying -H with @CO, so the default header line should be emitted at the start of the SAM, while the CO line should be after the SQ lines.

   ./bwa mem -H '@CO  ID:id' phix.fasta phix.fastq 
   [M::bwa_idx_load_from_disk] read 0 ALT contigs
   @HD VN:1.5  SO:unsorted GO:query
   @SQ SN:gi|9626372|ref|NC_001422.1|  LN:5386
   @CO ID:id

[jmarshall]

Superceded by #348, which completely refactors this function.

[nh13]

Closing in favor of #348 (more flexible) and in the hopes that this increases the chances the latter is merged!