feat(events): tensorize scheduling carriers

[TATP-233]

Summary

• Move EventManager interval timers to runtime-device float64 Torch tensors.
• Move reset-throttling step counters and trigger-once flags to device-resident int64/bool Torch tensors.
• Preserve env-owned NumPy RNG sampling for interval initialization, reset refresh, and retrigger sampling; this slice does not change the RNG stream.
• Keep the existing NumPy event-term API at the term boundary and update scheduling-carrier tests to assert Torch dtype/device.

Part of #1703.

Validation

• UNILAB_LOCAL_UNISIM=/home/user/ws/unilabsim/unisim uv run pytest -q tests/managers/test_event_command_metrics_recorder.py tests/envs/test_manager_based_rl_env.py tests/envs/mdp/test_events.py — 124 passed
• UNILAB_LOCAL_UNISIM=/home/user/ws/unilabsim/unisim uv run make check
• UNILAB_LOCAL_UNISIM=/home/user/ws/unilabsim/unisim uv run make test-all — 1832 passed, 26 skipped; benchmark imports 35/35 and 36/36

Remaining migration boundary

Event term functions and backend randomization transactions still consume NumPy payloads. Their device-resident transaction migration remains tracked by #1703.