[envpool] clean up dev Python dependencies and stabilize state buffers

[Trinkle23897]

Summary

• Problem: the dev dependency set still carried legacy pins and packages such as protobuf<=4.20.0, box2d-py, and mujoco_py, while the phase-4 cleanup surfaced a StateBufferQueue::Wait() path that could block forever when the background stock-buffer queue was temporarily empty.
• Scope: refresh the dev and benchmark Python requirements for the current Linux/JAX/Gymnasium stack, update the Box2D correctness baseline and Bazel deps, refresh the bazel_rules_qt pin, and harden the state-buffer fallback path. This does not change the release requirements or take on a broader bazel_clang_tidy migration.
• Outcome: Linux dev requirements resolve without the legacy Box2D/MuJoCo Python packages, Box2D correctness tests run against current Gymnasium baselines, and full make bazel-test is stable again on dev.

This diff modernizes the phase-4 dev dependency surface and closes the queue-preallocation flaky that showed up during the full Linux test sweep.

Technical Details

• Approach: remove obsolete Python packages from the dev-only requirements, refresh the Linux lockfile from a real dev resolution, keep the minimal runtime extras still required by Bazel tests, and make the state-buffer preallocation path fall back to synchronous allocation instead of blocking on an empty stock queue.
• Code pointers:
  • third_party/pip_requirements/requirements-dev.txt: removes the protobuf cap, drops box2d-py / mujoco_py, updates opencv-python-headless, and keeps imageio for the MuJoCo Gym test runtime.
  • third_party/pip_requirements/requirements-dev-lock.txt: refreshed from a Linux dev environment to match the new dependency surface.
  • envpool/box2d/box2d_correctness_test.py: switches the baseline side of the Box2D space checks to Gymnasium's current v3 env IDs.
  • envpool/box2d/BUILD: depends on box2d and gymnasium directly so Bazel sandboxes still see the right runtime packages.
  • envpool/core/circular_buffer.h: adds a non-blocking TryGet() helper.
  • envpool/core/state_buffer_queue.h: uses TryGet() and falls back to a synchronous StateBuffer allocation when the background stock queue is temporarily empty.
  • envpool/workspace0.bzl: refreshes the bazel_rules_qt pin.
  • BUILD: exposes .clang-tidy via srcs for the current aspect consumer.
• Notes: threadpool and gym3_libenv were checked and already match upstream head, so there is no diff for them. I also probed a newer bazel_clang_tidy, but that turned into a much larger lint/toolchain migration and is intentionally not included here.

Test Plan

Automated

• brix ssh dev -C -- 'cd /root/code/envpool && USE_BAZEL_VERSION=8.6.0 bazelisk test //envpool/core:state_buffer_queue_test --config=test --test_output=errors --nocache_test_results --runs_per_test=10': passed (10/10).
• brix ssh dev -C -- 'cd /root/code/envpool && make bazel-test': passed (30/30).

Suggested Manual

• python -m venv /tmp/envpool-phase4 && source /tmp/envpool-phase4/bin/activate && pip install -r third_party/pip_requirements/requirements-dev.txt: verify the dev dependency set resolves cleanly on Linux.
• USE_BAZEL_VERSION=8.6.0 bazelisk test //envpool/box2d:box2d_correctness_test --config=test --test_output=errors: verify the Gymnasium Box2D baseline continues to line up with EnvPool.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 3b5a6d7162

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".