Refactor file uris to make possible skip normalization

[jneira]

• After Normalize file paths before converting to Uri's #205 normalization could happen up to 2 times: in filePathToUri and toNormalizeUri, the last one to ensure that file NormalizedUri's from other sources (like Aeson) are always normalized (and fix a bug in hie)
• Moreover, if client code already has ensured that filepaths are normalized it suffer the penalty of two additional normalizations:
  • It happened in ghcid: Fix performance regression introduced by filepath normalisation ghcide#303 with a notable perfomance degradation (it is called many times)
  • So it was neccesary to inline the code of filePathToUrito avoid the extra normalization
• To avoid this, filePathToUri should remove the normalization, moving it to normalizeUri and maybe to another function filePathToNormalizedUri :: FilePath -> NormalizedUri

[cocreature]

One option would be to move NormalizedFilePath to haskell-lsp. Then we can provide a safe NormalizedFilePath -> NormalizedUri function that can skip normalization.

[alanz]

@cocreature sounds good to me. Would anyone like to make a PR?

[jneira]

@alanz i am afraid i dont know the ghcide codebase enough to implement @cocreature suggestion, does worth make a first small step like the described one in my comment?

For the more advanced refactoring i would enforce (maybe optionally) that file paths are absolute (NormalizedAbsoluteFilePath?), cause file uri's must be absolute by the standard and afaik hie and ghcide already use them.

[jneira]

I am working on this.