Tracking issue for deprecation lint `legacy_derive_helpers`

[petrochenkov]

This is a tracking issue for deprecating code like this:

#[trait_helper] // This helper attribute is used before it introduced by `#[derive(Trait)]` below.
#[derive(Trait)]
struct S;

Code like this works for historical reasons, but attribute expansion works in left-to-right order, so we have to use some hacks and "look into the future" to resolve trait_helper, but that future may never actually happen in cases like

#[trait_helper]
#[remove_all_derives] // removes the `#[derive(Trait)]` below
#[derive(Trait)]
struct S;

, so the resolution in this case is unreliable.

Steps

• A deprecation lint for this case is added in expand/resolve: Turn #[derive] into a regular macro attribute #79078 as warn-by-default.
• Change the lint to report in dependencies Mark all deprecation lints in name resolution as deny-by-default and report-in-deps #143929
• Switch to a hard error