Refactor extraction and caching of CabalInfo

[brandonchinn178]

Broken off from #1033

This change has the following improvements:

1. Force user to handle missing stanza information (e.g. dependencies) when the cabal file does not mention a source file
   • This would make it much harder to have the same regression as #1032 somewhere else
2. Make CabalInfo return all info for the cabal file, without reference to the source file
   • Simplifies parseCabalInfo as just "parse cabal file", not "parse cabal file + extract info relevant to given source file"
   • IMO this is clearer, given the names of the types and functions.

[mrkkrp]

I'm happy with the first commit, which I adjusted a bit to apply the new helper to both caching of cabal files and .ormolu files, however I don't think we can easily merge something that changes the signature of refineConfig because it changes the API, which we are supposed to keep backwards compatible now as per c81320d8a1069674c17fa5ee52b64c18604f84de.

[brandonchinn178]

Ah sure, I could make the changes backwards compatible. Give me a second

[brandonchinn178]

@mrkkrp how's that?

Also, there's a weeder failure. Maybe weeder.dhall should be updated to include the Ormolu module as a root?

[mrkkrp]

Eh, I like splitting the code in two branches (old and new) even less. Let's apply d91bbd064511fee913bd348fc413ba5a820e8927 and keep the rest on the shelf for a while until we decide to release 0.8.0.0 at which point we can make a clean break.

[brandonchinn178]

Sure up to you