Currently, benchcab does not have extensive coverage in its unit tests.
Parts of the code that have poor test coverage are dependent on system
interaction (e.g. environment modules, running subprocess commands (SVN,
build scripts, executables, PBS).

This change refactors and adds unit testing for these parts of the code
base.

Move the code from the benchcab.run_cable_site module to the
benchcab.task module. This is done so that we reduce coupling between
modules that use fluxnet tasks. This also has the benefit that we can
test the two modules easily by sharing the same test setup functions. We
move the code from the benchcab.run_comparison module to
benchcab.task module for the same reasons.

Unit tests now check exception messages and standard output produced by
benchcab for both non-verbose and verbose modes. **Beware:** standard
output messages have been updated in some areas. These changes will need
to be updated in the documentation.

Refactor the default build so that environment modules are loaded via
python instead of writing module load statements into the temporary
build script. This is done to reduce the complexity of the unit tests.

The stderr output from subprocess commands are now redirected to stdout
so that stderr is suppressed in non-verbose mode.

Fixes #58 #22 #86

This change refactors the code to be more object oriented so that we can
better support mocking via dependency injection rather than resorting to
the `unittest.mock.patch` function. This allows us to write unit tests
that are simpler and that preserve the API layer (as opposed to using
`unittest.mock.patch` which breaks the API layer).

Fixes #102

Co-authored-by: Claire Carouge <[email protected]>