fix: align off-policy logger timing and header

[TATP-233]

Summary

• start off-policy elapsed-time and ETA accounting only after the replay buffer first becomes ready for learner work, excluding collector startup, MuJoCo chunk tuning, and initial replay warm-up
• align the completed-run training_wall_time_sec and Perfetto learner/training_e2e window with the same boundary for single-GPU runners
• move the algorithm, task, and current iteration into the panel title (🚀 UniLab Off-Policy Training | FastSAC | G1WalkFlat | iter 5000/5000), remove iteration from the compact statistics row, and remove the normal | Training suffix while preserving error statuses
• keep Reward component labels such as penalty action rate, penalty ang vel xy, and penalty orientation on one line in narrow terminals
• render Env Step detail timings as dim cyan children with right-aligned mirrored tree connectors (12.5ms ─┤, 1.0ms ─┤, 0.5ms ─┘), keeping the tree aligned across different numeric widths
• cover the logger output and timer boundary across legacy single-GPU, double-buffer, and multi-GPU runners

Linked Work

• Issue: none
• Milestone: none

Validation

• make check
• uv run pytest -m "not slow"
• Additional task-specific validation listed below

Commands actually run:

uv run pytest tests/algos/test_offpolicy_logger.py -q
uv run pytest tests/algos/test_offpolicy_logger.py tests/algos/test_offpolicy_runner_unit.py -q
uv run pytest tests/algos/test_offpolicy_logger.py tests/algos/test_offpolicy_runner_unit.py tests/utils/test_experiment_tracking.py -q
make test-all

make test-all: 1634 passed, 29 skipped, 267 deselected, 1 xfailed; benchmark smoke test passed (2 platform-optional MLX entries skipped).

Impact

• Backend impact: none
• Platform impact: both
• Training effect expected: no; training behavior is unchanged, while terminal and run-summary timing accounting changes

Artifacts

• W&B: none
• benchmark result: benchmark smoke test passed
• video / screenshot: none
• ONNX / checkpoint: none

Checklist

• Added or updated tests where needed
• Updated docs if behavior or workflow changed (not needed; terminal presentation only)
• Linked the driving issue (none provided)
• Noted any follow-up work explicitly (none)