Better ignore doc comment

[hhugo]

Fix #2408
Floating doc comment were not ignored when using --no-comment-check, only attached doc-comment were.

This change can result in empty Ptop_def [] which are now ignored while mapping.

For the record, I've tested this PR on the all sources inside opam-respository

find ../all-sources -name "*.ml" -type f -o -name "*.mli" -type f | parallel --progress _build/default/bin/ocamlformat/main.exe --enable-outside-detected-project -q --no-comment-check --no-version-check --ignore-invalid-option > /dev/null

I don't see any Ast changed bug anymore

[Julow]

I'm not sure we should ignore docstrings while checking. They are part of the AST and loosing or moving one would be catastrophic. Perhaps the normalization function could be improved instead ?

What's your use case for --no-comment-check ?

[hhugo]

The purpose of --no-comment-check is to ignore comments and doc-comments !

Here is what the ocamlformat doc says

Control whether to check comments and documentation comments. Unsafe to turn off. May be set in .ocamlformat. The flag is set by default.

The flag was introduced for debug/dev purposes so that one case focus on Ast changed errors. I use this in #985

[Julow]

Doc comments are part of the AST and I don't consider them problematic like comments. I think doc comments issues are of higher priority than other Ast changed errors.
I'd prefer to fix these first rather than to ignore them.
What do you think?

[hhugo]

Please looks at what's done currently. Please look at the documentation for this option. This option is not supposed to be used by regular users. It currently already ignores some doc-comments. This PR just make the behavior consistent.

I don't really understand your pushback on this developers option.

This change allows to run ocamlformat on files triggering warning 50 and spot Changed ast without the noise caused by comments.

[hhugo]

Also see PR introducing the feature #259

[hhugo]

This PR is restoring the feature introduced in #259 that was altered at some point later

[hhugo]

The regression was introduced by #1672
The following statement was invalid

or double checking things like ignore_doc_comments (checked by the attributes method anyway), maybe these were considered to be optimizations but I didn't observe any slowdown;

[Julow]

Alright, you convinced me it's useful, let's merge.