feat(sim): switch replay state to dynamic joints

[juelg]

Summary

• replace the replayed sim state payload behind the existing sim_state flow with dynamic joint state
• keep the original replay wrapper, replay script, and replay tests, but adapt them to the new schema-aware mechanism
• warn when replay only partially matches the current scene instead of failing silently
• keep replay working across added or removed fixed scene elements and on fr3_dual_arm

Notes

This PR is stacked on top of feat/sim-state-record-replay and is meant to review only the follow-up that swaps the underlying replay state mechanism.

The replay path still uses the original:

• SimStateObservationWrapper
• scripts/replay_sim_trajectory.py
• python/rcs/sim_state_replay.py
• python/tests/test_sim_state_record_replay.py

But the Sim state implementation underneath now uses named dynamic-joint schema plus flat values, so scene-only changes do not break replay.

Validation

• make stubgen
• make pyformat
• make cppformat
• python -m pip install -e .
• pytest -q python/tests/test_sim_state_record_replay.py
• pytest -q python/tests/test_multi_sim.py python/tests/test_sim_state_record_replay.py

[juelg]

Closing this stacked PR per request. I’m moving the work back to the original master-based PR instead.