In `#[allow(...)]` completion, prefer lints that are firing

A common action (especially when using clippy) is to observe reported lint and #[allow()] it. Therefore, it would be useful if, when completing with the cursor is inside an #[allow()], the ranking preferred the names of lints for which at least one diagnostic has been emitted within the syntax tree to which the #[allow()] attribute applies.

It probably makes sense to generalize this to any combination of attribute and diagnostic severity.

• allow and expect would prefer current warnings and errors
• deny and forbid would prefer current warnings
• warn would prefer current errors

I took a look at implementing this, but I didn't find a way to fetch info about current diagnostics from inside ide_completion::completions::attribute::complete_known_attribute_input.

Start at ide_completion::completions::attribute::complete_known_attribute_input and investigate how diagnostics for the attribute's syntax tree could be made available there. Define the severity-to-attribute mapping described in the issue, then verify that lint completion ranking prefers currently emitted diagnostics for allow, expect, deny, forbid, and warn.