feat(tasks): unify G1 motion tracking identity

[TATP-233]

Summary

• Unifies the G1 motion tracking registry identity to G1MotionTracking across PPO/APPO/SAC/FlashSAC. The single semantic identity now registers the supported Manager-Based backend factories; algorithm, backend, DR, observation, and tensor-runtime differences remain owned by Hydra task YAML.
• Removes the redundant G1MotionTrackingSAC registration and updates configs, tests, benchmarks, fixtures, migration matrix, and bilingual docs accordingly.
• Restores semantic FlashSAC reward telemetry. MotionRewardPack and MotionPenaltyRewardPack now emit raw per-component columns; their YAML target terms own canonical names and weights, so RewardManager publishes motion_global_root_pos/ori, motion_body_pos/ori/lin_vel/ang_vel, action_rate_l2, joint_limit, and undesired_contacts without executing the targets twice.
• Updates the FlashSAC owner fingerprints for the packed reward contract transition.
• User-facing training impact: new SAC/FlashSAC run roots use logs/fast_sac/G1MotionTracking/ and logs/flash_sac/G1MotionTracking/. Existing G1MotionTrackingSAC runs still load by passing an absolute algo.load_run path.

Linked Work

• Issue: driving work tracked in this PR; upstream terminal-logger split/width issue: Off-policy terminal: separate IPC diagnostics from losses and add width-aware layout unilabsim/unilab_rl#86
• Parent roadmap: none
• Roadmap declared base: none
• Milestone: none
• Base branch: main

Validation

• make test-all passed on the final local head before this PR was created or updated
• Additional task-specific validation listed below

Commands actually run:

uv run pytest -q tests/managers/test_core_managers.py tests/config/test_g1_motion_tracking_owners.py tests/tasks/test_flashsac_owner_contract.py tests/tasks/test_motion_term_parity.py tests/tasks/test_production_registry_closeout.py tests/tasks/test_package_boundary.py tests/scripts/test_train_scripts.py tests/scripts/test_train_script_configs.py tests/scripts/test_check_docs.py tests/scripts/test_g1_flashsac_backend.py tests/utils/test_experiment_tracking.py
# 168 passed, 137 deselected

make check
# passed

make test
# 1823 passed, 42 skipped, 488 deselected

make test-all
# passed: format/lint/mypy/pyright, 1823 tests with coverage, benchmark import smoke 34/34 module and 35/35 script

# Manual runtime smoke, FlashSAC + MuJoCo, num_envs=2, one zero-action step:
# reward info keys: action_rate_l2, joint_limit, motion_body_ang_vel,
# motion_body_lin_vel, motion_body_ori, motion_body_pos,
# motion_global_root_ori, motion_global_root_pos, motion_joint_pos,
# motion_joint_vel, motion_reward_pack, undesired_contacts

Remote CI route:

• Base main: current-head CI will be monitored after PR creation.

Impact

• Backend impact: mujoco / motrix (registry/config coverage also includes mjwarp, genesis, and newton owners; runtime smoke run on MuJoCo)
• Platform impact: Linux (validation environment)
• Training effect expected: yes — unified task/run identity and restored FlashSAC reward diagnostics; reward equations and aggregate values remain unchanged

Artifacts

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

Checklist

• Added or updated tests where needed
• Updated docs if behavior or workflow changed
• Linked the driving issue
• Noted any follow-up work explicitly

Follow-up: terminal logger metric classification and width-aware layout are owned by unilab-rl and tracked in unilabsim/unilab_rl#86.

[TATP-233]

Rebased onto current main after unilab-rl 1.4.6 SAC naming landed and resolved the two documentation conflicts. Revalidated final head 1a060ea0 locally with uv sync --extra mujoco --extra uni_rl, focused tests, make check, make test, and make test-all. Current-head CI is green: ruff lint/format, mypy, pyright, benchmark-smoke, test (ubuntu-slim), and Sphinx.