Propagate run:once errors to all dependents

[edrevo]

Motivation

Let's say we have this Gremlins-inspired Taskfile.yml

Just for context, Gremlins cannot be fed or bathed after midnight or they will become evil.

version: '3'
run: once
tasks:
  ensure-before-midnight:
    cmds:
      - fail if it's after midnight
  feed-gremlins:
    deps:
      - ensure-before-midnight
    cmds:
      - feed the gremlins
  bathe-gremlins:
    deps:
      - ensure-before-midnight
    cmds:
      - bathe the gremlins

If it is after midnight and we run task feed-gremlins the task fails. Same thing for task bathe-gremlins. However, if we run task bathe-gremlins feed-gremlins --parallel then one of the tasks will run and the other will be skipped, and we end up with an evil Gremlin.

Which one depends entirely on timing issues.

❯ task bathe-gremlins feed-gremlins --parallel -C 0
task: [ensure-before-midnight] exit 1
task: [feed-gremlins] sleep 2
task: Failed to run task "bathe-gremlins": task: Failed to run task "ensure-before-midnight": exit status 1

❯ task bathe-gremlins feed-gremlins --parallel -C 0
task: [ensure-before-midnight] exit 1
task: [bathe-gremlins] sleep 2
task: Failed to run task "feed-gremlins": task: Failed to run task "ensure-before-midnight": exit status 1

Change

This PR propagates not only the completion status of dependent tasks, but also the exit status of the tasks. That means that if task A has task B as a dependency, it will never run if task B failed (this is currently not the case).

I have taken the conservative route and added the new behavior behind a CLI flag, although I would personally define this as a bug given the fact that which task gets run and which doesn't is non-deterministic.