ci: split pytest suites and schedule environment smoke tests

[wlgys8]

Closes #94

Summary

• split pytest CI into parallel fast and numba suites for PRs and pushes to main/stable
• schedule the full suite, including all-environment subprocess smoke tests, nightly and on manual dispatch
• register and apply slow, numba, and integration pytest markers
• keep the complete WBT Numba module together so first-use kernel compilation is paid by the dedicated suite rather than the fast suite
• upload JUnit test results and report the slowest test phases

Validation

• prek run --files .github/workflows/README.md .github/workflows/tests.yml pyproject.toml motrix_envs/tests/test_wbt_numba.py test/test_all_envs.py
• fast suite: 1007 passed, 2 skipped in 78.33s
• Numba suite: 37 passed in 59.25s
• pytest --collect-only --strict-markers for fast, Numba, and integration selections
• git diff --check

The full local pytest command remains unchanged and continues to run all tests.

[Copilot]

Copilot review overview

🔵 Needs a closer look

Renaming the single pytest check into pytest (fast)/(numba)/(full) changes merge-gating status checks and requires a human to update branch-protection settings to keep required checks enforced.

Review effort: Balanced
Findings: None

What changed in this PR

This PR restructures the tests.yml CI workflow to reduce day-to-day feedback latency while preserving Numba-runtime and full-environment coverage (closes #94). It introduces pytest markers to tier tests, splits PR/push runs into parallel fast and numba suites, and defers the expensive all-environment subprocess smoke test (test_all_demos) to a nightly/manual-dispatch full suite. Documentation in the workflow README explains the tiers, local commands, and guidance for future Numba environments.

Changes:

• Register slow/numba/integration markers in pyproject.toml and apply them (test_all_demos → slow+integration; whole test_wbt_numba.py module → numba, two long tests → slow).
• Replace the single CI job with an event-driven matrix: fast/numba on PR/push, full on schedule/dispatch; add JUnit upload and --durations=20 with --strict-markers.
• Add a nightly cron, per-event concurrency group, per-suite Numba cache keys, and expanded README documentation.

File	Description
test/​test_all_envs.py	Marks the all-environment smoke test as slow+integration so it only runs in the full/nightly suite.
pyproject.toml	Registers the slow, numba, and integration pytest markers used for CI tiering.
motrix_envs/​tests/​test_wbt_numba.py	Module-level numba marker keeps WBT kernels in one suite; adds slow to the two expensive tests.
.github/​workflows/​tests.yml	Event-driven suite matrix, nightly schedule, per-suite caches, JUnit artifacts, and --strict-markers.
.github/​workflows/​README.md	Documents the new suite tiers, local commands, marker semantics, and future-Numba guidance.

---

💡 Add a code-review agent skill or configure MCP servers for context-aware, tailored reviews. Learn more in the docs.

[Copilot]

Copilot review overview

🔵 Needs a closer look

The CI/marker changes are low-risk, but the accompanying rewrites to core Numba kernel cloning, fingerprinting, and compilation-cache logic are subtle and correctness-critical, warranting final human review.

Review effort: Balanced
Findings: None