Support loading structural variants

[iskandr]

Currently we assume that every variant is a focal change with a nucleotide alt sequence. This is only a subset of the VCF spec and doesn't allow us to load variants such as:

#CHROM  POS ID  REF ALT QUAL    FILTER  INFO    FORMAT  synthetic.challenge.set4.normal synthetic.challenge.set4.tumour
1   959526  MantaDEL:42:0:1:0:0:0   T   <DEL>   .   MinSomaticScore END=962016;SVTYPE=DEL;SVLEN=-2490;IMPRECISE;CIPOS=-86,86;CIEND=-65,66;SOMATIC;SOMATICSCORE=18   PR  12,0    8,5
1   4932289 MantaDUP:TANDEM:263:0:1:0:0:0   A   <DUP:TANDEM>    .   PASS    END=4943947;SVTYPE=DUP;SVLEN=11658;SOMATIC;SOMATICSCORE=73  PR:SR   24,0:60,0   14,6:28,6
1   8411313 MantaDUP:TANDEM:461:0:1:0:0:0   A   <DUP:TANDEM>    .   PASS    END=8419870;SVTYPE=DUP;SVLEN=8557;SOMATIC;SOMATICSCORE=55   PR:SR   11,0:20,0   5,5:14,8
1   10856998    MantaINV:563:0:1:0:0:0  C   <INV>   .   PASS    END=10864731;SVTYPE=INV;SVLEN=7733;CIPOS=0,5;CIEND=-5,0;HOMLEN=5;HOMSEQ=CCCCC;INV3;EVENT=MantaINV:563:0:1:0:0:0;SOMATIC;SOMATICSCORE=79;JUNCTION_SOMATICSCORE=32    PR:SR   9,0:15,0    2,5:10,4

Though we probably can't do much to predict the protein sequence for most of these variants we should still be able to load them and distinguish them from focal variants.

[armish]

Surfaced because of this error within Cycledash: hammerlab/cycledash#840