chore: update unilab-rl to 1.3.0

[TATP-233]

Summary

• Raise the standard and ROCm unilab-rl requirements from 1.2.1 to the released 1.3.0, updating both lockfiles.
• Consume the off-policy collector-ready handshake and inference-timeout forwarding fixes from unilab_rl 1.3.0.
• Add the now-required TD3 training.inference_request_timeout_sec owner field and update the SAC/FlashSAC comments to the post-handshake timeout semantics.

Linked Work

• Issue: roadmap: g1_motion_tracking multi-backend enablement — newton CUDA graph, genesis cold-start, isaac worker fixes #1604
• Parent roadmap: roadmap: g1_motion_tracking multi-backend enablement — newton CUDA graph, genesis cold-start, isaac worker fixes #1604
• Roadmap declared base: main
• Milestone: none
• Base branch: main
• Upstream release: https://github.com/unilabsim/unilab_rl/releases/tag/v1.3.0

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 lock
uv lock --check
# Activated pyproject.rocm.toml / uv.rocm.lock temporarily, then:
uv lock
uv lock --check
uv run pytest -q tests/algos/test_offpolicy_double_buffer_runner.py tests/scripts/test_train_scripts.py::test_offpolicy_flashsac_g1_motion_tracking_task_composes
uv run --no-sync python -c 'from importlib.metadata import version; import uni_rl; print(version("unilab-rl")); print(uni_rl.__file__)'
make check
make test-all
git diff --check

Results:

• Focused test: 24 passed, 4 deselected; installed unilab-rl reports 1.3.0.
• make check: Ruff, mypy, and Pyright passed; Pyright retains the existing optional Drake import warning.
• make test-all: 1491 passed, 27 skipped, 601 deselected; benchmark smoke passed 34/34 module mode and 35/35 script mode.

Remote CI route:

• Base main: current-head CI passed: https://github.com/Motphys/UniLab/actions/runs/35215080692 (all required jobs green).

Impact

• Backend impact: none
• Platform impact: Linux / macOS dependency profiles both resolved; no backend-specific runtime change
• Training effect expected: yes, off-policy collectors now use the collector-ready handshake and no longer spend the inference timeout on backend cold start

Artifacts

• W&B: none
• benchmark result: none needed for dependency compatibility
• 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