feat(multiplexer): Use single channel for muxer

[scarmuega]

The muxer loop uses a lock-free spin to iterate over available ingress channels to select the next payload to be muxed. This is not very efficient since we end up performing a lot of wasted cycles, even when using some backoff mechanism (such as the snooze function in the crossbeam crate).

This PR resolves the problem by relying on a single ingress channel, which provides a blocking mechanism using OS-primitives. This removes the need for a loop altogether. After some initial testing, we see that an example chainsync process against a local node maintains the same throughput (~1.5k blocks / sec) while drastically reducing the amount of required CPU (<4%).

The downside (?) is that now the muxer messages will processed in a FIFO fashion, instead of the randomized approach used in the actual implementation. This seems like a valid trade-off, possible even better for certain use-cases.