Increase granularity of watershed bypass

[ArtemSokolov]

Currently, instance segmentation models are accommodated via --s3seg-opts '--nucleiRegion bypass', which bypasses the watershed segmentation step. However, --s3seg-opts is applied universally, making it impossible to run probability map generators in parallel with instance segmenters. The current workaround is to do the parallelization by hand:

# Steps up to registration are in common
nextflow run labsyspharm/mcmicro --in exemplar-001 --stop-at registration

# Run probability map computation separately
nextflow run labsyspharm/mcmicro --in exemplar-001 --start-at probability-maps --probability-maps unmicst,ilastik

# ...from instance segmentation, which requires watershed bypass
nextflow run labsyspharm/mcmicro --in exemplar-001 --start-at probability-maps --probability-maps mesmer --s3seg-opts '--nucleiRegion bypass'

I propose to allow watershed bypass to be specified on a per-module basis. The most straightforward way is to introduce a new configuration setting bypass, which will be true for instance segmenters and false for probability map generators. Taking into account #317, example specs may look as follows:

[
  name      : 'unmicst',
  container : 'labsyspharm/unmicst',
  version   : '2.7.0',
  cmd       : 'python /app/unmicstWrapper.py --stackOutput --outputPath .',
  input     : ''
  channel   : '--channel'
  idxBase   : 1
  bypass    : false
]

[
  name      : 'mesmer',
  container : 'vanvalenlab/deepcell-applications',
  version   : '0.1.0',
  cmd       : 'python /usr/src/app/run_app.py mesmer --output-directory .',
  input     : '--nuclear-image'
  channel   : '--nuclear_channel'
  idxBase   : 0
  bypass    : true
]