Fix for the InDel dropping behaviour during consensus masking

[will-rowe]

This PR fixes the following problem reported by Nick:

For data where there is a large indel being reported in the VCF, the indel isn't being applied to the consensus as the region is part-covered by a mask and bcftools silently ignores it.

This PR does the following:

• adds test data, including some reads that produce an INDEL against the nCov reference
• adds a functional test to catch a case when an INDEL is currently being dropped by bcftools
• the above test required a bit of code shuffling in artic/pipeline.py, moving the parser to it's own function so that functional tests can call it easily (I also removed some unused imports)
• separates unit tests from functional tests by file extension (xxx_unit_test.py and xxx_medaka_test.py) and updates travis to reflect this
• updates minion.py so that input to longshot is tabix indexed, this was required for the tests to work as the subset of reads only covered a small portion of the reference genome
• updates make_depth_mask.py to fix the reported INDEL dropping problem:
  • removes existing function that calls samtools depth
  • replace with a call to mpileup via pysam
  • add option to include query deletions in reference position counts, such that the depth of coverage for any genome position covered by a deletion is incremented
  • update the CLI for the artic_make_depth_mask command so that you can replicate the old behaviour (using pysam) with --ignore-deletions