Make tcx.visibility() work for items in the local crate

[jyn514]

This also changes rustc_metadata to use tcx.visibility() for record!ing metadata.

See #77820 (comment) for the history.

r? @petrochenkov

[jyn514]

The important commit is 3a88f79, the first one just switches lots of places from ty::Visibility::from_hir(def_id) to tcx.visibility(def_id). I recommend reading it with whitespace changes hidden. The current compile error is

error[E0004]: non-exhaustive patterns: `Ty(_)`, `TraitRef(_)`, `Binding(_)` and 8 more not covered
   --> compiler/rustc_privacy/src/lib.rs:254:25
    |
254 |     let hir_vis = match tcx.hir().get(hir_id) {
    |                         ^^^^^^^^^^^^^^^^^^^^^ patterns `Ty(_)`, `TraitRef(_)`, `Binding(_)` and 8 more not covered

I don't know how to handle these.

[petrochenkov]

@jyn514
Every record!(self.tables.visibility[...] <- RIGHT_HAND); in encoder.rs should use tcx.visibility(def_id) on the right hand side.
The body of fn visibility should be equivalent to those right hand sides combined.
That's what I previously mentioned in #77820 (comment) and #77820 (comment).

Touching rustc_privacy::def_id_visibility is not a goal right now, it may do different things from tcx.visibility(def_id) and I don't remember the exact details, I'll audit it later.
(If some pieces of rustc_privacy::def_id_visibility are obviously equivalent to those from tcx.visibility(def_id) based on encoding.rs, then tcx.visibility(def_id) can be used in implementation of rustc_privacy::def_id_visibility, but it's not a goal to replace the whole rustc_privacy::def_id_visibility with tcx.visibility(def_id).)

[jyn514]

I think I did it right this time :) It passes src/test/ui at least.

[jyn514]

Let me know if I should switch more places to using tcx.visibility() instead of calculating it themselves, I've only done rustc_metadata so far.

[petrochenkov]

So I ended up fixing this properly and deduplicating visibility calculations in all three places in which they are performed - resolve (which can't use queries), metadata encoding, and privacy.

I'm still debugging, but if I finish that work today or at least tomorrow, I'll close this PR in favor of it.

[petrochenkov]

Closing in favor of #78077.