Should StageError::Internal be fatal?

[onbjerg]

Per #453 there are two "categories" for the variants of StageError: fatal and non-fatal.

Fatal errors stop the pipeline completely because they are assumed to be irrecoverable (e.g. db corruption), and non-fatal errors result in the pipeline retrying the stage again.

Currently, the only fatal variants in StageError are database related. There is one catch-all variant called StageError::Internal and it is unclear if this should be fatal or not: stages may want custom errors for informational purposes that are not fatal, but some errors from e.g. the standard library might be fatal (currently I don't think any stage emit errors from the standard library).

[gakonst]

Let's split theInternal error in 2 variants Fatal and Recoverable which box any kind of error so that the pipeline can know whether to stop or retry.

[onbjerg]

Done here: 9a66b07