Introspection api

[vnermolaev]

This PR improves monitor-based actor introspection #26

• Consolidates ActorMonitor state into per-actor locked stats instead of a single shared structure.
• Expands monitor-derived metrics with dispatch, delivery, handled, error, overflow, and derived queue-depth data.
• Refactors ActorController’s select! loop so branch selection and work execution are separated, simplifying scoping around step() execution. Without this wiring of on_step_enter/on_step_exit created a lot of borrowing issues.
• Wires the previously unused on_step_enter and on_step_exit hooks into the runtime.
• Uses those step callbacks in ActorMonitor to derive a best-effort actor state from StepAction.

Given the current design, this best-effort actor state is the best option we have right now without bloating the code with additional controller-side reporting.