feat: allow modules to be marked as errors

[markdalgleish]

This adds an error module type so that Node built-ins can be marked as errors if they're present in the final output.

I've described this in a comment in the code, but this is done by scanning the generated code so that unused modules can be part of the module graph. This way it's only an error if they're not tree shaken.

This is useful in the context of Remix because we have Node built-ins in the module graph that may never be used on the client. This is due to the way route files contain both server and client code. In this case the fallback option is super useful because I can break the Remix browser build if an unpolyfilled module is detected in the output, whereas our v2 branch currently falls back to an empty module which is a runtime error. This approach is much nicer for us.

Status and versioning classification:

• Code changes have been tested and working fine, or there are no code changes
• This PR changes the library's interface (methods or parameters added)

[imranbarbhuiya]

Changes LGTM but why write option need to be false?

[markdalgleish]

@imranbarbhuiya Because you can't read the contents of outputFiles in the onEnd callback unless write is set to false.

[imranbarbhuiya]

released in v1.5.0