Also update the argparse help text for the --discordant-vcf flag to
refer to a regular VCF file, instead of a gzipped VCF file.

Merge testing into master

See merge request !6

In the VCF files generated by GATK, DP is an integer, while the GQ field
holds a float.

 ##FORMAT=<ID=DP,Number=1,Type=Integer
 ##FORMAT=<ID=GQ,Number=1,Type=Float

When GATK emits all sites, homref sites get the RQC annotation instead
of the QC annotations. For vtools, these two annotations are used
interchangeably, since they both contain genotype likelihoods.

See the detailed explantation at
https://gatkforums.broadinstitute.org/gatk/discussion/5115/haplotypecaller-genotypegvcfs-and-vqsr-for-alternatives-in-dbsnp

When a vcf file contains calls for all specified positions, the ALT field will
be empty if there is only a reference call in that position. This commits adds
tests for this case.

To fix this, we had to rewrite the logic used when comparing variants.
As a side effect, 'allele_no_call' is now only counted for partial calls
when they occur in the call set. The test case
test_partial_positive_no_call was updated to reflect this change.

Also fixed an error where the known and called variants were switched
when they were being passed to parse_variants.

The actual genotypes are the same between gatk.vcf and
gatk_ref_alt_changed.vcf, but the REF, ALT and corresponding genotype
calls differ. This is to test if the conversion of genotype call
(e.g. 1/3) to the actual genotype (e.g. A/T) is working as intended.

Since it is not clear how we should handle non-diploid calls, we check
if variants are haploid/polyploid and error out if this is the case.

Before, we would iterate over decomposed called variants until we found
the variant where the REF and ALT matched the one in the positive vcf.
However, this meant that sites where the called vcf had different ALTs
could not be processed.

To resolve this, we had to drop support for decomposed variants.

Since it is not clear how we should handle phased calls, we check if the
variants are phased and error out if this is the case.

Previously we used cyvcf2 gt_types, which indicate whether a site is
hom_ref, heterozygous, hom_alt or unknown. A limitation of this approach
is that it only makes sense when the two vcf files have the same REF and
ALT alleles for a given position.

If this was not the case, vtools would silently skip the variant.
However, even when the ALT alleles are different between two vcf files,
the actual genotypes can still be compared. This was implemented in this
commit, which also fixes issue #8.

There are two ways in which a variant can be no call. The variant can be
present in the vcf file with a ./. genotype, or the variant can be
missing from the vcf file altogether. This commit adds a test case to
verify that no call sites are counted properly when they are missing
from the called vcf file.

Sites that are missing from the positive vcf file do not count towards
alleles_no_call, since we do not care about variants which are not
present in the positive vcf file.

cyvcf2 assigns 'no call' sites a DP and QC of -1, which caused the
statistics to be off since we checked for 'no call' sites (i.e. gt=3)
after all other checks.

The check for 'no call' sites was moved up to make sure only true sites
get to the QC/DP checks. Furthermore, the QC/DP checks are counted
independenly, since a site can fail multiple checks.

The test cases were updated to reflect these changes, and now all pass.