Currently in dagger, there is no way to influence the way each pipeline is run. They either all works the first time or the whole plan (i.e. dagger up) fails.

We need a way to have more control on the pipelines execution. It should be possible to:

• Declaring a task in a way that it's expected to fail (eg. useful for negative unit tests, expected to fail in some cases)
• Declaring a behavior for retry (eg. a task to retry every 1 second during 10 seconds total)
• Trigger another task only if a previous one fail