Addresses #32409 (comment)

Pros: Visibilities can be applied to tuple struct fields, only two symbols of look-ahead are required.

struct S(pub(foo::bar) u8);

parses as struct with a pub(restricted) field of type u8.

Cons: Some tuple structs stop parsing

struct S(pub (foo::bar::<u8>, foo::baz));

We start parsing pub (foo::bar as a visibility and fail somewhere in the middle.

Why it is not so bad:

• The breakage seems to be rare, at least rustc itself doesn't contain any affected code, even in tests. Crater run is needed.
• The workaround is simple - surround the type with parentheses - ((foo, bar)). ( isn't a path start token, so ((foo, bar)) parses as a type. Alternatively, a type alias can be used for (foo, bar).
• It doesn't affect macros. If a macro accept types we can pass a tuple (foo::bar::<u8>, foo::baz) to it without fear of syntactic conflicts even if it's preceeded by pub in the macro.

macro_rules! m {
    ($t: ty) => (struct S(pub $t);)
}

m! { (foo::bar::<u8>, foo::baz) } // OK

Alternatives:

• Parse pub (path) TYPE with hacks. Always parse what goes after pub as a type (but not $ty nonterminal) (unless it's (crate) or $vis) and try to reinterpret it as a visibility if it's not followed by , or ).
  Reinterpretation fails if the parsed path is not TyKind::Path wrapped into TyKind::Paren or contains type parameters.
• Use lookahead of arbitrary length, parse the whole path first and decide what to do after that. A path can be an actual visibility or a tuple's first element, or an operand of type sum, etc.
• Support only unambiguous pub(crate) (and $vis if Add a vis matcher to macro_rules!. rfcs#1575 is accepted) in tuple structs. This is essentially this PR other way round - everything that starts like a type parses as a type. However workarounds are wordy if pub(restricted) is actually wanted.
• Use some other syntax until the current one is de-facto stabilized (I, personally, like pub@path, but would prefer to keep the current syntax).
• Do not support pub(restricted) in tuple structs. This just looks like a surrender.

[breaking-change], needs a crater run before proceeding.
r? @nikomatsakis