Issue #340: Add benchmark env smoke test to CI

[Sammy-Dabbas]

Closes #340.

What this does

Adds a run-benchmarks job to commit_checks.yml plus a small benchmarks/smoke_test.py driver. The job builds via the existing .github/actions/build action (Python 3.11, pip-options: all,rllib, mirroring check-doc-build) and then constructs one representative environment from each benchmark module -- nadir_science, aeos_single, and rso_inspection -- and steps each 3 times.

Design choices

• Smoke test, not full benchmark. benchmark.py --total_timesteps 0 still spins up Ray and builds a full PPO model, so the smoke script bypasses Ray/RLlib entirely and exercises only environment construction and the PettingZoo step loop. Breakages in benchmark env definitions fail in minutes instead of requiring a training run.
• env_args carries the RLlib-only episode_data_callback/satellite_data_callback keys that ConstellationTasking.__init__ does not accept; the script strips them the same way _constellation_tasking_env_creator does.
• The script runs with working-directory: benchmarks because the benchmark modules use top-level imports (from benchmark import BenchmarkEnv).
• Single Python version (3.11) rather than the full pytest matrix to keep CI cost low.

Verification

Basilisk is a heavy native build, so locally I validated the workflow YAML, ruff/isort cleanliness, and traced every invocation against the actual signatures (ConstellationTasking kwargs, PettingZoo agent-keyed step dicts). The end-to-end run happens in this PR's own CI via the new job. Happy to adjust scope (e.g. more envs per module or a different Python version) if you'd prefer.