New rustdoc lint to respect -Dwarnings correctly

[poliorcetics]

This adds a new lint to rustc that is used in rustdoc when a code
block is empty or cannot be parsed as valid Rust code.

Previously this was unconditionally a warning. As such some
documentation comments were (unknowingly) abusing this to pass despite
the -Dwarnings used when compiling rustc, this should not be the
case anymore.

Todo List

• Declare lint.
• Document lint (file: src/doc/rustdoc/src/lints.md).
• Use lint in rustdoc (file: src/librustdoc/passes/check_code_block_syntax.rs, maybe others).
• Write tests.
• Refactor things.
• Squash into one commit.

---

Previously was #79816 but I broke the submodules on my side and it was less trouble fixing it with a new PR that cherry-picked the changes.

Fix #79792.

@rustbot label T-rustdoc A-lint
r? @jyn514

[petrochenkov]

Can lints reported by rustdoc be defined entirely in rustdoc without touching rustc_lint and other compiler crates?
Clippy does this somehow.

I've been seeing numerous lints of uncertain quality being added to rustdoc in the recent year without any process from the compiler or lang team side, while in theory new lints require lang team RFCs or FCPs.
If they are added by rustdoc team, can they at least be offloaded entirely into rustdoc?

[poliorcetics]

It would certainly be both easier to maintain and less buggy to have everything concentrated in rustdoc but I don't know if it is feasible, I'm especially not knowledgeable about how lints work in the compiler to tell.

I'll try to make something work and report my findings.

[jyn514]

@poliorcetics please don't make that change here, I would prefer to do it consistently and not piecemeal.

Can lints reported by rustdoc be defined entirely in rustdoc without touching rustc_lint and other compiler crates?

I would like to do that but haven't gotten around to it. The main technical issue is that the old lint names will have to be deprecated and renamed; they can't be deleted, because it's not possible to register built-in lints from out-of-tree (in other words, the new lints will have to be named rustdoc::lint, not lint directly).

There's also some debate about which lints should belong to rustdoc and which should be part of rustc: #77364

in theory new lints require lang team RFCs or FCPs.

I didn't know that. Is that written down somewhere?

[poliorcetics]

Closing this in favour of #79816, see there for more infos.