resume after crash.

[airtonix]

Gasket steps (and any step in a generic task runner) should be idempotent, but in reality users are free to use whatever i.e rm -rf / o.0.

So, instead the pipeline should support resuming from last failed step.

A way of dealing with this that I've seen before is :

1. by storing a hash of the pipeline manifest,
2. creating a temporary $PWD/gasket-progress.json file which stores current progress of the pipeline execution
3. on error, bail. show error explaining the presence and purpose of the progress file
4. allow the user to fix errors. and run gasket resume
5. continue execution from last failed step specified in $PWD/gasket-progress.json.
6. bail if pipeline manifest source (gasket.json, package.json, etc) changes (since progress wouldn't match)