[envpool] migrate Bazel Qt rules in-tree

[Trinkle23897]

Summary

• Problem: EnvPool still depends on the archived bazel_rules_qt repository even though the repo only needs a very small local Qt wrapper for Procgen, and the current branch also needed a small box2d typing fix to keep lint moving.
• Scope: Replaces the external bazel_rules_qt archive with repo-local Qt configure/build files used by Procgen, and includes a narrow box2d_correctness_test.py mypy fix. It does not change threadpool, gym3_libenv, or bazel_clang_tidy pins.
• Outcome: Bazel no longer fetches the archived Qt ruleset, Procgen continues to build against the local Linux Qt install, and box2d_correctness_test.py no longer trips mypy on scalar/shape comparisons.

This keeps the Qt dependency surface in-tree and removes a dead external Bazel rules dependency without changing Procgen's linked Qt libraries.

Technical Details

• Approach: Add a minimal Linux-only third_party/qt repository rule that resolves the local Qt include directory and exposes the qt_core/qt_gui libraries EnvPool actually uses; then wire WORKSPACE and workspace1.bzl to those local rules.
• Code pointers:
  • third_party/qt/qt_configure.bzl: minimal local Qt repository rule that finds the Qt headers and templates local_qt.bzl
  • third_party/qt/qt.BUILD: repo-local qt_core / qt_gui Bazel targets used by Procgen
  • WORKSPACE: rewires @qt to the in-repo Qt BUILD file and drops the external Qt toolchain registration
  • envpool/box2d/box2d_correctness_test.py: swaps two assert_allclose calls for assertEqual to satisfy mypy on shape/discrete-cardinality checks
• Notes: upstream justbuchanan/bazel_rules_qt is archived and the previous pin was already at upstream head, so this is a replacement rather than a version bump.

Test Plan

Automated

• make buildifier: pass
• make addlicense: pass
• make mypy: pass
• USE_BAZEL_VERSION=8.6.0 bazelisk test //envpool/procgen:procgen_env_test //envpool/procgen:procgen_test --enable_workspace --noenable_bzlmod --config=test --test_output=errors --nocache_test_results on dev-0: pass
• make bazel-test on dev-0: 30 / 30 tests pass
• make lint on dev-0: in progress; flake8, py-format, clang-format, cpplint, buildifier, addlicense, mypy, docstyle, and spelling have passed, and only the long-running clang-tidy tail remains

Suggested Manual

• make lint on a Linux devbox: confirm the remaining clang-tidy tail finishes cleanly on the exact tree in this PR

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

Reviewed commit: 7316cdc541

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".

[chatgpt-codex-connector]

Track BAZEL_RULES_QT_DIR in the repository rule

If a developer sets or changes BAZEL_RULES_QT_DIR after the first fetch, Bazel can keep reusing the cached @local_config_qt repository because qt_autoconf does not declare that environment variable as an input. Reading repository_ctx.os.environ directly here means the generated local_qt.bzl can stay pinned to a stale header path until the user manually cleans or syncs, which makes switching/fixing Qt installations unexpectedly fail.

Useful? React with 👍 / 👎.

[chatgpt-codex-connector]

Add the configured Qt libdir to the link flags

When BAZEL_RULES_QT_DIR points at a non-system Qt installation (for example /opt/Qt/...), this target still links with bare -lQt5Core/-lQt5Gui flags and never adds that prefix's lib directory. In that setup the headers resolve because new_local_repository is rooted at the include tree, but the final link still searches only the default system paths and fails to find the Qt shared libraries.

Useful? React with 👍 / 👎.