*: reduce all "deny" and "forbid" lints to "warn"

[benesch]

Lints at the deny/forbid level can be very frustrating when developing,
because you're forced to comply with the lint even when you're still
prototyping, just to get your code to compile.

This commit downgrades all these lints to "warn", to optimize for the
development experience. In CI, these lints are still enforced at the
"deny" level by our Clippy configuration, which upgrades all warnings to
hard errors.

The one glitch here is that forbid used to prevent submodules from
re-enabling the forbidden lint. I propose that we solve this socially,
by making it taboo to add #![allow(missing_docs)] unless you've gotten
buy-in from the overall owner of that area of the codebase.

[antifuchs]

Seeing as how I already do this for dataflow in #7690, I'm ok with this! The "no turning the level down" check does hurt a bit to lose, but this might be an interesting skunkworks project (;

[frankmcsherry]

I propose that we solve this socially, by making it taboo to add #![allow(missing_docs)] unless you've gotten buy-in from the overall owner of that area of the codebase.

We could remove the four occurrences, and then lint for it in CI.

[umanwizard]

yes, I love this

[benesch]

I propose that we solve this socially, by making it taboo to add #![allow(missing_docs)] unless you've gotten buy-in from the overall owner of that area of the codebase.

We could remove the four occurrences, and then lint for it in CI.

Yeah, the big sticking point there is https://github.com/MaterializeInc/materialize/blob/main/src/pgrepr/src/oid.rs. I feel like enabling missing_docs there would be more pain than it's worth? I could be convinced, though, by the simplicity of the resulting lint. I sent #7693 to remove one of the other allow(missing_docs) lints. That just leaves two other instances which should be fixed by adding docs.