fix: integrate PR #3315 review feedback

- envs(robotwin): default `observation_height/width` in
  `create_robotwin_envs` to `DEFAULT_CAMERA_H/W` (240/320) so they
  match the D435 dims baked into `task_config/demo_clean.yml`.
- envs(robotwin): resolve `task_config/demo_clean.yml` via
  `CONFIGS_PATH` instead of a cwd-relative path; works regardless
  of where `lerobot-eval` is invoked.
- envs(robotwin): replace `print()` calls in `create_robotwin_envs`
  with `logger.info(...)` (module-level `logger = logging.getLogger`).
- envs(robotwin): use `_LazyAsyncVectorEnv` for the async path so
  async workers start lazily (matches LIBERO / RoboCasa / VLABench).
- envs(robotwin): cast `agent_pos` space + joint-state output to
  float32 end-to-end (was mixed float64/float32).
- envs(configs): use the existing `_make_vec_env_cls(use_async,
  n_envs)` helper in `RoboTwinEnvConfig.create_envs`; drop the
  `get_env_processors` override so RoboTwin uses the identity
  processor inherited from `EnvConfig`.
- processor: delete `RoboTwinProcessorStep` — the float32 cast now
  happens in the wrapper itself, so the processor is redundant.
- tests: drop the `TestRoboTwinProcessorStep` suite; update the
  mock obs fixture to use float32 `joint_action.vector`.
- ci: hoist `ROBOTWIN_POLICY` and `ROBOTWIN_TASKS` to job-level
  env vars so the task list and policy aren't duplicated across
  eval / extract / parse steps.
- docker: pin RoboTwin + CuRobo upstream clones to commit SHAs
  (`RoboTwin@0aeea2d6`, `curobo@ca941586`) for reproducibility.

Co-Authored-By: Claude Opus 4.7 (1M context) <noreply@anthropic.com>

docker/Dockerfile.benchmark.robotwin

@@ -30,6 +30,9 @@ ENV NVIDIA_DRIVER_CAPABILITIES=all \
 # The nightly base is CUDA -base (no compiler, no Vulkan loader). CuRobo's
 # `pip install -e .` runs nvcc, and SAPIEN renders via Vulkan — add both.
 USER root
+# Pinned upstream SHA for reproducible benchmark runs. Bump when we need
+# an upstream fix; don't rely on `main` drift.
+ARG ROBOTWIN_SHA=0aeea2d669c0f8516f4d5785f0aa33ba812c14b4
 RUN apt-get update \
     && apt-get install -y --no-install-recommends \
          cuda-nvcc-12-4 cuda-cudart-dev-12-4 \
@@ -37,7 +40,8 @@ RUN apt-get update \
     && mkdir -p /usr/share/vulkan/icd.d \
     && echo '{"file_format_version":"1.0.0","ICD":{"library_path":"libGLX_nvidia.so.0","api_version":"1.3.0"}}' \
        > /usr/share/vulkan/icd.d/nvidia_icd.json \
-    && git clone --depth=1 https://github.com/RoboTwin-Platform/RoboTwin.git ${ROBOTWIN_ROOT} \
+    && git clone https://github.com/RoboTwin-Platform/RoboTwin.git ${ROBOTWIN_ROOT} \
+    && git -C ${ROBOTWIN_ROOT} checkout ${ROBOTWIN_SHA} \
     && chown -R user_lerobot:user_lerobot ${ROBOTWIN_ROOT} \
     && apt-get clean && rm -rf /var/lib/apt/lists/*
 USER user_lerobot
@@ -52,9 +56,11 @@ RUN uv pip install --no-cache --no-build-isolation \
         "git+https://github.com/facebookresearch/pytorch3d.git@stable"
 
 # CuRobo — NVlabs motion generator; TORCH_CUDA_ARCH_LIST must be set or the
-# build aborts on an empty arch list.
+# build aborts on an empty arch list. Pinned SHA for reproducibility.
+ARG CUROBO_SHA=ca941586c33b8482ed9c0e74d60f23efd64b516a
 RUN cd ${ROBOTWIN_ROOT}/envs \
-    && git clone --depth=1 https://github.com/NVlabs/curobo.git \
+    && git clone https://github.com/NVlabs/curobo.git \
+    && git -C curobo checkout ${CUROBO_SHA} \
     && cd curobo \
     && TORCH_CUDA_ARCH_LIST="7.0;7.5;8.0;8.6;8.9;9.0" \
        uv pip install -e . --no-build-isolation --no-cache