Concurrent execution policies for pure representable machines.
[ExclusiveExecutor] directly returns outputs to a caller that serializes
turns through exclusive access. [SerializedExecutor] provides
run-to-completion turns: it queues inputs
and does not advance the next transition until the preceding output handler
returns. [LinearizedExecutor] advances inputs immediately under its lock,
then dispatches already-ordered outputs. The latter policy is appropriate
only when transition linearization may precede completion of earlier work.