Rework lock sharing between Selector instances

[MaxiBoether]

Right now, we pre-allocate 64 locks and assign each pipeline a lock based on their ID. This is (just as in #308), because sub-processes (the gRPC server instances) cannot allocate new locks on demand from a manager. This would need access to the manager in the subprocess. We would like to have 1 lock per Selector instance instead of a fixed pre-allocated amount of locks.

This might require some rework of how we do multiprocessing in the Selector, which can be done in conjunction with #308.