Drop dependency on ghc-source-gen

[TeofilC]

ghc-source-gen is tightly coupled to the GHC AST. This is an inherently unstable interface that can and does change with each version of GHC.

This makes GHC upgrades difficult for a project that depends on proto-lens and therefore ghc-source-gen, since each release requires a lot of CPP to be added.

I'm wondering if it would be possible to move away from using ghc-source-gen. I think the safety gains from using the actual AST are quite low, while the costs from coupling are quite high. A lot of the complexity comes from exactprint annotation which don't seem to be used by this library at all.

I think using a pretty printing library to generate Haskell ASTs would be much easier to maintain. Note that the unlike the AST, the textual surface syntax is a stable interface. Alternatively we could use the template-haskell AST.

I would be happy to put some time into implementing this if this seems like a good idea.

[blackgnezdo]

@judah is probably the best person to opine on this kind of change even though he didn't have much time recently to dedicate to the project. The idea of building on top of GHC APIs sounded appealing when we started. With each intermediate release the amount of conditional code has been growing, so maybe with this new information in mind we can reconsider.

[erikd]

@TeofilC I would love to see you replace ghc-source-gen in this project.

I have had a look at the compile issues with ghc-source-gen building with ghc-9.12 and they are incredibly hairy and IMO require a deep understanding of the GHC AST and how it changes from major GHC version to version.