FIX: Verify first word of _cmd in dependency check

[effigies]

Changes proposed in this pull request

In the event that an interface._cmd has spaces in it - i.e., pre-set flags in addition to the command name - the which check will fail. This PR uses _cmd.split()[0] to use only the command up to the first space when checking for the command.

A command of the form python /some/path/some_script will merely check the existence of python, and not /some/path/some_script, but that would not be correctly handled by the existing check. To my knowledge, we do not have any such nodes at this time, so this is theoretical, but we may want to consider a more robust long-term solution, but I don't believe that should hold up this fix.

Fixes #1235.

Documentation that should be reviewed

None.

Pull-request guidelines:

• I have read the guidelines for contributions.
• I understand that my contributions will not be merged unless the work is
  finished (i.e. no [WIP] tag remains in the title of my PR) and tests pass.
• The proposed code follows the
  coding style,
  to the extent I understood them (and I will address any comments raised by the PR's reviewers in this regard).
• (Optional) I opt-out from being listed in the .zenodo.json file.

[effigies]

@mgxd Are you able to check build a Singularity image off of this branch, prior to release? It would be nice to have a check.

The issue should arise in Docker, as well, given the same inputs. The error criterion, I believe, should be whether a QwarpPlusMinus interface is instantiated in our workflow, and has nothing to do with Singularity.

[oesteban]

I feel pretty confident this will fix the issue. Let's go ahead. @mgxd please reopen if I'm wrong. Please, let us know that it worked otherwise.

[mgxd]

@effigies @oesteban this is a good short-term solution, but should probably be reverted once the relevant interfaces are fixed and released upstream.

One case that comes to mind is wb_command <command>, in which the could vary between workbench versions. This way of checking will always assume the command works (as long as wb_command is found in the PATH)

[effigies]

Reverting would break wb_command <command> even further. We don't pretend to check whether the arguments will work, just whether we can hit the executable.