Detect and don't report non-breaking renaming re-exports.

[obi1kenobi]

Only use importable paths to match items across versions.

Declared names (e.g. Foo in pub struct Foo;) are subject to change in renaming pub use and pub type statements (e.g. pub use Foo as Bar), so relying on them to match items across versions means we miss renames. The importable paths provided by trustfall_rustdoc v0.8.2+ account for this, and list all valid paths under which the item can be imported. These paths are the only way we can match items across versions.

The above logic change needed to be applied to all currently-written lints, and will need to also apply to all future lints as well. Heads-up for @SmolSir @tonowak @staniewzki @mgr0dzicki.

The new logic is also somewhat more expensive, since it increases the search space and eliminates some of the pruning that filtering on the name could get us. This is unfortunate, but unavoidable at the moment — we don't want to sacrifice correctness for speed. I'll focus on perf improvements next and I'm confident we can win back all the lost performance and then some.

Fixes #288, #215, #202.
Fixes #42.

[obi1kenobi]

Still TODO here:

• add all crates that had these false-positives into the GitHub Actions regression suite
• measure the perf impact of the new lint changes

I think the impact won't be spectacular on most crates, but still better to measure and make an explicit decision based on data. Hope is not a strategy.

[epage]

Didn't get into too much of the details but glad to see this done

[obi1kenobi]

An overnight perf run didn't show any significant perf changes one way or the other on even the biggest crates.