refactor(workspace): standardise on async_trait, drop hand-rolled BoxFuture

[varex83agent]

Closes #613.

Reworked after review: instead of splitting traits between native async-fn-in-trait and #[async_trait], every trait with an async method now uses #[async_trait].

• Traits that hand-rolled BoxFuture-returning methods purely to wrap already-async inherent methods (consensus::wrapper::Consensus, dkg::bcast::RegisteredMessage) now declare plain async fn.
• priority's non-trait send_receive is a plain async fn instead of returning a boxed future.
• The attribute is spelled the same way workspace-wide: bare #[async_trait] plus use async_trait::async_trait, which was already the majority form (29 vs 6). The six qualified #[async_trait::async_trait] uses were normalised.
• cli's DiskTestTool — the one native-AFIT trait in the tree — joins them, retiring the lone #[allow(async_fn_in_trait)] and its comment.
• core's duplicate BoxFuture alias is dropped in favour of futures::future::BoxFuture.

Out of scope and untouched: the libp2p ConnectionHandler state-machine BoxFutures, and the subscriber callback type aliases owned by #606.

Note

This inverts issue #613's stated acceptance criterion ("#[async_trait] remains only on dyn-used traits") in favour of the uniformity asked for in review. Worth editing or closing #613 to match.

Co-Authored-By: Bohdan Ohorodnii 35969035+varex83@users.noreply.github.com

[emlautarom1]

Not sure we gain much from switching between native and async_trait definitions. I'd suggest using async_trait everywhere that we have a trait with an async method for consistency. Be consistent on the attribute usage: either #[async_trait::async_trait] or #[async_trait]. Also remove the comments detailing how the trait is being used, it can and will get stale and provides no value at all to the reader.

The replacement of the custom BoxFuture with the one from futures makes sense and looks good.

[emlautarom1]

Drop these kind of comments.

[varex83agent]

Done — removed.

[varex83agent]

Thanks — reworked to go the other way. Everything with an async trait method is now #[async_trait]:

• Reverted the native-AFIT conversions on AggSigDB and FTransport; they're back on #[async_trait].
• Kept the conversions of the hand-rolled BoxFuture-returning trait methods (consensus::wrapper::Consensus, dkg::bcast::RegisteredMessage) — they're plain async fn under #[async_trait] now, so the Box::pin(async move { ... }) wrappers over already-async inherent methods are gone.
• Consistent spelling: bare #[async_trait] with use async_trait::async_trait everywhere. That was already the majority (29 vs 6), so the six qualified #[async_trait::async_trait] uses were normalised — including three files outside the original diff (priority/{component,prioritiser}.rs, priority/tests/prioritiser_test.rs).
• Dropped all the comments explaining how each trait is used.
• Kept the BoxFuture → futures::future::BoxFuture swap.

One judgement call worth a look: cli's DiskTestTool was the tree's only native-AFIT trait, carrying an #[allow(async_fn_in_trait)] and a four-line comment about it. To make "async_trait everywhere" actually true I converted it too, which adds async-trait to crates/cli (already a workspace dep). Happy to drop that hunk if you'd rather keep the diff to the files originally touched.

Note this inverts #613's acceptance criterion ("#[async_trait] remains only on dyn-used traits") — the issue probably wants editing or closing to match.

Gates: fmt, clippy -D warnings, test --workspace --all-features (57 suites), cargo deny all green.