Make job completion order deterministic, improve tests

[cottsay]

When multiple jobs complete in the same "tick" of the parallel execution loop, the done_futures set can result in different colcon return codes depending on which one happens to get processed first. A more desirable behavior is to prefer the return code of the job which was started first.

While it is currently unlikely to see multiple jobs finish in the same tick due to strict concurrency pre-limiting, this becomes much more common under dynamic scheduling and resource-guarded execution.

This change also includes an additional test for the parallel executor, and switches all of the job identifiers to use consistent identifiers, where they were previously using different values in the overall job dictionary from the identifier used on the job objects themselves.

Assisted-by: Gemini 3.5 Flash

[codecov-commenter]

Codecov Report

✅ All modified and coverable lines are covered by tests.
✅ Project coverage is 97.38%. Comparing base (94dfeff) to head (7b5e008).

Additional details and impacted files

@@            Coverage Diff             @@
##           master      #42      +/-   ##
==========================================
+ Coverage   96.73%   97.38%   +0.65%     
==========================================
  Files           4        4              
  Lines         153      153              
  Branches       41       41              
==========================================
+ Hits          148      149       +1     
  Misses          1        1              
+ Partials        4        3       -1     

☔ View full report in Codecov by Harness.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[knmcguire]

I'll review this so add me as reviewer @knmcguire

[knmcguire]

Should we also have a test on the first addition you made?

Claude Opus 5.1 generated this:

class ImmediateJob(Job):

    def __init__(self, identifier, rc):
        super().__init__(
            identifier=identifier, dependencies=set(), task=None,
            task_context=None)
        self.rc = rc

    async def __call__(self, *args, **kwargs):
        return self.rc


def test_simultaneous_completion_prefers_first_started():
    extension = ParallelExecutorExtension()
    args = SimpleNamespace(parallel_workers=2)

    # repeat: set iteration order varies run to run, so a single
    # pass would only catch the bug ~50% of the time
    for _ in range(50):
        jobs = OrderedDict()
        jobs['early'] = ImmediateJob('early', 3)
        jobs['late'] = ImmediateJob('late', 7)
        rc = extension.execute(args, jobs, on_error=OnError.continue_)
        assert rc == 3, f'expected first-started rc 3, got {rc}'

I've tested it out on the current master and the patch, and indeed it passes on your fix.

FYI, I had to use Claude Opus 5.1 to check out a couple of assumptions for the review as here but I tested off of the changes myself.

[cottsay]

Should we also have a test on the first addition you made?

Sure! Added verbatim in 7b5e008.

[knmcguire]

Looks good! thanks!