Assign an AttrStyle to all HIR attributes

[dtolnay]

Fixes #142649.

Prior to #135726, all hir::Attribute used to have a "style" (either AttrStyle::Outer or AttrStyle::Inner, for #[…] and #![…] respectively). After that refactor, hir::Attribute's fn style(&self) -> AttrStyle began to panic if called on certain builtin attributes like #[deprecated].

I looked into changing this method to fn style(&self) -> Option<AttrStyle> where builtin attributes would not have a style. But ultimately the HIR pretty-printer needs to print all these attributes, and needs to print them in either outer or inner position, so at least every attribute included in -Zunpretty=hir definitely needs a style, including builtin attributes.

I looked into storing every attribute's original AttrStyle in either rustc_attr_data_structures::AttributeKind or hir::Attribute. This does not make sense because there is a many-to-one correspondence between AST attributes and HIR attributes. Two different AST attributes with different style can get parsed into the same single HIR attribute. I have added an example of this in a comment.

Since the HIR pretty-printer chooses to print all of these builtin attributes as outer attributes, and we need to be able to know the style of every expression-level attribute in order to correctly handle expression precedence during pretty-printing and diagnostics, in this PR I have made hir::Attribute::style return AttrStyle::Outer instead of panicking for builtin attributes.

This solution fixes several rustc_hir_pretty bugs associated with not being able to know the style of HIR attributes. For example, some attributes used to get duplicated as both outer and inner when printed, as you can see in the tests.

@jdonszelmann @fmease

[rustbot]

r? @oli-obk

rustbot has assigned @oli-obk.
They will have a look at your PR within the next two weeks and either review your PR or reassign to another reviewer.

Use r? to explicitly pick a reviewer

[jdonszelmann]

I expect to, within a few weeks or so, convert all attributes to be parsed. As such, I'm not sure the changes in this pr are very useful as this function will likely be deleted soon since it would effectively always return Outer. Only tool attributes might retain the ability to have an AttrStyle, though I think it'd be acceptable to always print those as outer too.

[dtolnay]

Closing in favor of #142776.