Change: replace async_trait with RPITIT

[hadronzoo]

Completes #959 by removing async_trait and updating the add_async_trait proc macro to add Send bounds when the singlethreaded feature is not enabled. This also removes the requirement for annotating trait implementations with a proc macro.

For example, the original RaftRuntime trait:

#[add_async_trait]
pub(crate) trait RaftRuntime<C: RaftTypeConfig> {
    async fn run_command<'e>(&mut self, cmd: Command<C>) -> Result<Option<Command<C>>, StorageError<C::NodeId>>;
}

is transformed into the following when the singlethreaded feature is not enabled:

pub(crate) trait RaftRuntime<C: RaftTypeConfig>: Send {
    fn run_command<'e>(&mut self, cmd: Command<C>) -> impl std::future::Future<Output=Result<Option<Command<C>>, StorageError<C::NodeId>>> + Send;
}

This PR is covered by existing unit tests.

---

This change is 

[schreter]

Nice!

From my side, it looks good.

+reviewer:@drmingdrmer

It would be also nice to measure in a local microbenchmark (so the network is excluded) the performance before and after the change - it namely significantly reduces the number of (expensive) allocations for futures and enables much better inlining across async functions.

Reviewed 34 of 34 files at r1, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @drmingdrmer)

[drmingdrmer]

It's pretty nice! It's a very clean approach! Thank you!

Reviewed all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @hadronzoo)

[drmingdrmer]

Nice!

From my side, it looks good.

+reviewer:@drmingdrmer

It would be also nice to measure in a local microbenchmark (so the network is excluded) the performance before and after the change - it namely significantly reduces the number of (expensive) allocations for futures and enables much better inlining across async functions.

Reviewed 34 of 34 files at r1, all commit messages.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on @drmingdrmer)

make bench_cluster_of_3 shows there is a performance improvement of 6%. :D
1,016 op/ms to 1,077 op/ms

[drmingdrmer]

@hadronzoo
I manually merged your PR into main branch with a fix of an unused crate async_trait in crate rocksstore-v2.
And it looks like github just closed this PR because of the change to main :(

[hadronzoo]

@drmingdrmer thank you for accepting this change!