This took me longer than expected.

I've fixed another bug along the way: LINE pragma in a disabled region no longer leads to an AST change error.

Some thoughts and remarks about this MR:

• Ormolu.Processing.Common was moved to Ormolu.Processing.Disabling, as it only contains definitions related to disabling. This is analogous to Ormolu.Processing.Cpp.
• The disabled regions could perhaps be pretty-printed in Ormolu.Parser.Result.prettyPrintParseResult (currently I just show them). I didn't find this necessary.
• The preprocessing part is ugly. I've found it difficult to make it look better using the current structure of the code. I thought about restructuring it as a composition, more or less like: preprocess = removeLinePragmas . removeDisabledRegions . maskCPP . removeShebang, or using state to make it more readable. I decided that it's a big change, and I'd rather hear your opinion first.
• The test cases in tests/Ormolu/Parser/ParseFailureSpec.hs could perhaps be specified using srcSpans instead of strings. The current implementation relies on the text representation of the srcSpan, which may change (but I believe it is unlikely to change), It is, at least in part, a result of me being lazy, but it is also short and readable. If you think this should be rewritten, let me know.

Thanks for your effort on this PR! We chose to go with a slightly different approach (formatting the non-disabled snippets individually, see #773). Your work here is very appreciated, and your new tests are actually reused (of course, you are attributed as a coauthor).

@amesgen Ok, thanks! I haven't read through #773 yet, but the solution here wasn't very clean, so I'm pretty sure you've done better. I strongly agree with the decision to remove support for #601 . It was a major pain point while I was working on this PR. I've found this behaviour inconsistent and considered it a misfeature. Perhaps I should have complained to Mark about it instead of making workarounds.