Summary

This PR fixes the "off-by-one" recording error in the experimental DataRecorder (introduced in #3145) by correctly aligning the recorded data with its actual chronological timestamp.

Bug / Issue

In the new Discrete Event Simulation (DES) architecture, the DataRecorder exhibits a bug where it permanently captures the "past" state of the model, but tags it with the "future" timestamp.

The root cause lies in the execution order of the DES engine:

1. The engine pops an event for the next time slice (e.g., $t=1.0$).
2. The engine updates the simulation clock (self.time = event.time).
3. The engine executes the event (event.execute()).

Because model.time is an Observable, updating the clock in step 2 instantly broadcasts a CHANGED signal. The DataRecorder reacts to this signal and takes a snapshot of the agents before they are actually allowed to execute their logic in step 3.

Previously, the recorder was tagging this snapshot with the new time signal. This meant the initial state was being recorded and labeled as $t=1.0$, the $t=1.0$ state was labeled as $t=2.0$, and so on, permanently shifting the data one step out of alignment.

Implementation

This PR resolves the timestamp misalignment by extracting the pre-transition time from the Observable signal by changing the timestamp assignment in _on_time_change to use current_time = signal.additional_kwargs.get("old").

Fixes #3297