Run snaps in worker thread or Node.js VM to get easier to read errors

[Mrtenz]

Right now if a snap fails to evaluate in SES, you get a error with a stack trace that is quite useless. I propose that we evaluate snaps twice. First, we run it in a Node.js VM or worker thread, ..., to catch any runtime errors that show up when loading the snap, and then we run it again with SES, to catch any compatibility errors.