fix: bind InternalRPC and heartbeat tasks to the provided runtime

[PopClom]

connect_uri_with_runtime accepts an explicit runtime, but InternalRPC::run and the heartbeat task are spawned via async_rs's runtime.spawn(), which checks Handle::try_current() before the stored handle. Since the runtime is typically constructed via tokio_current() — which captures a Handle but does not hold an owned TokioRT — self.runtime is None, and try_current() always wins when called from an async context, ignoring the handle that was explicitly provided.

When the caller's runtime has a shorter lifetime than the connection — e.g. a #[tokio::test] runtime sharing a connection pool across tests — InternalRPC::run dies when the test ends, silently breaking publish confirmations.

Fix: bump async-rs to 0.8.2, which checks the stored handle before Handle::try_current().

[Keruspe]

Nice catch, thanks!

I'd rather keep the internal_rpc start where it is though and not add more not-really-related things to the io loop.
This brought to light a bug in async-rs though, which is now fixed in 0.8.2.

Could you drop the move to io_loop; keep your test and bump the dependency of async-rs to 0.8.2 to confirm everything is ok this way?

[PopClom]

Makes sense, thanks! Bumped async-rs to 0.8.2 and the test passes with just that change 🙌