refactor: extract shared rl runtime

[annedranowski]

Summary

• Extracts the shared Spherogram RL environment and PPO load/train runtime from the two target notebooks.
• Updates unknotting_curr_ver and unknotting_connect_sum_experiments to use the shared runtime module.
• Keeps this PR stacked on refactor/shared-invariants so the RL runtime diff can be reviewed separately.

Validation

• Parsed edited notebooks for Python syntax.
• Ran PYTHONPATH=src .venv/bin/python -m compileall -q src tests.
• Ran PYTHONPATH=src .venv/bin/python -m pytest tests/test_rl_runtime.py tests/test_invariants.py tests/test_parsing.py tests/test_config.py tests/test_checkpoints.py tests/test_paths_workbook_targets.py.