Rework the CQueue implementation

[NicolasT]

Instead of using CQueue from distributed-process, provide an implementation which is simpler and tailored to our needs.

• There's no need for built-in timeout support (cfr. troupe: add after in favor of receiveTimeout #36)
• There's no need for built-in non-blocking support (cfr. troupe: add after in favor of receiveTimeout #36)
• It's likely a good idea to make it work in STM so Receive is non-atomic w.r.t. signals #29 is not an issue

This is meant to be a long-running PR to experiment on the implementation. The intent is for the implementation to be correct, not necessarily fast. Once it's correct, we can further work on its performance.

The tests for the implementation include some experiments with DejaFu to ensure the implementation produces consistent results and doesn't deadlock. Cfr. #17.

[NicolasT]

There's clearly some bug in the current implementation, since using it "for real" causes some tests to lock up.

I might first rework Process so it can (internally) live in MonadConc, such that some DejaFu tests can ensure it works as intended.