Feature gate the `must_not_suspend` lint

[camelid]

Feature-gate the must_not_suspend lint (the attribute is already feature-gated) as #![feature(must_not_suspend_lint)]. This will have to be done before 1.57.0 is released (since that's when the lint will land).

cc #83310 (comment)

[camelid]

Instructions:

• Apply the changes to compiler/ in Feature gate the non_exhaustive_omitted_patterns lint #89428 to the must_not_suspend lint.
• Add test showing that #![warn(must_not_suspend)] emits a stability error.
• Add #![feature(must_not_suspend_lint)] to existing tests of must_not_suspend.

[guswynn]

I have a pr ready for this, but are we sure we want to? Having people get a lint for mutexguard use across await downstream, but only allowing nightly to annotate new types seems useful to me?

[guswynn]

oh wait, the lint still triggers, its just the feature gate on the using lint directives! seems fine to me!

[Mark-Simulacrum]

The lint should likely be allow by default unless users can disable it somehow in a stable way (otherwise it's a real annoyance, needing e.g. allow(warnings) to get rid of it). The main reason to feature gate + allow by default it initially is to make sure the decision to stabilize surface area is intentional, whereas it sounds like here we did so more or less by accident.

[guswynn]

@Mark-Simulacrum that is covered by this pr: #89787

I believe both landing is what we want for now!

[eholk]

@rustbot label +AsyncAwait-triaged