Previously the following expressions

    ID="@testid"
    ID="rs123"
    ID!="rs123"

would not match the ID string "rs123;rs456" as expected.

Resolves #2190

Resolves #1484

For example, these variants should not be removed with `-d exact`

    POS=1 ALT=<DEL> SVLEN=-1000
    POS=1 ALT=<DEL> SVLEN=-2000
    POS=1 ALT=<DEL> SVLEN=-3000

Resolves #2182

The bug was introduced with the output format change in 1.19
which replaced the DC section with DCv2
https://github.com/samtools/bcftools/commit/983e610ab

Fixes #2180

The command `bcftools norm -m+` will not work when the merged
genotypes have different ploidy. This commit attempts to fix that,
but note that it is experimental and I am having doubts if it should
be supported at all, given it may interfere with partially missing
genotypes produces by atomization

Resolves #1962

We move this to gcc instead.

This doesn't change the test harness at all.  It's simply extra
documentation and scripts to aid with large scale evaluation of
mpileup changes.

The user-format output of `bcftools query` and `bcftools +split-vep` prints
column indices by default, e.g. "#[1]CHROM". This can be now suppressed by
giving the option multiple times, `-HH`

Resolves #2152

Resolves #2160

see https://github.com/tecnickcom/variantkey

Resolves #2157

Resolves #2153

similarly to symbolic <DEL.*> alleles added previously (7040c101)

Resolves #2145

For example, in the two cases below the field 'STR' from the -a file is required to match
the INFO/TAG in VCF. In the first example the alleles REF,ALT must match, in the second
example they are ignored. The option -k is required to output also records that were not
annotated:

    bcftools annotate -a ann.tsv.gz -c CHROM,POS,REF,ALT,SCORE,~STR -i'TAG={STR}' -k in.vcf
    bcftools annotate -a ann.tsv.gz -c CHROM,POS,-,-,SCORE,~STR     -i'TAG={STR}' -k in.vcf

Resolves #2151

This needs more work, didn't realize that the effect of the -c option is to
update the target VCF as well, so can't be used simply for matching. For
that, probably a new prefix will need to be introduced, but that's already
confusing as it is, for example ~ has double meaning.

This reverts commit e9bff3fb. Switch to
branch bt-2151-annotate-ext to see this code.

For example, in the two cases below the field 'STR' from the -a file is required to match
the INFO/TAG in VCF. In the first example the alleles REF,ALT must match, in the second
example they are ignored. The option -k is required to output also records that were not
annotated:

    bcftools annotate -a ann.tsv.gz -c CHROM,POS,REF,ALT,SCORE,STR -i'TAG={STR}' -k in.vcf
    bcftools annotate -a ann.tsv.gz -c CHROM,POS,-,-,SCORE,STR     -i'TAG={STR}' -k in.vcf

Resolves #2151

Red Hat's ExtUtils::Embed returns CLAGS & LDFLAGS that produce a
PIE position-independent executable. However libhts.a and the rest
of bcftools's *.o files are not compiled as PIC, so linking fails
as these other objects use relocations that are invalid for PIE.

While bcftools could link against libhts.so and compile the rest of
its objects with -fpic, the logistics are non-trivial. So it's easier
to omit the redhat-hardened-cc1 and redhat-hardened-ld specs that set
up for building a PIE executable. Fixes #1322.

It stated "Missing the --fa-ref option", but the option is "--fasta-ref".

Given --write-index now takes an optional argument, and optional long
options are --long-opt=arg and short options are -larg, I chose to
also accept -l=arg for the sake of consistency and ease of
documentation.  The standard -larg still works too.

If we wish to stick strictly to the normal conventions, then this is a
trivial change in version.c (and a search and replace in the
documentation).

Also fix formatting bug in bcftools merge man page section leading to
a lot of underlined text.

Fixes #2139

This was previously segmentation faulting as not being a string means
it didn't set the key field. We now report it as a parse error.

Previously only the last file had the index written.

Note this is the only subcommand which defaults to writing TBI for
VCF.gz.  Everything else defaults to CSI.  I'm unsure why this is, but
I haven't changed it in this PR.