Add Paper Problem 1 upstream reference generation

[bernalde]

Summary

• Add benchmarks/paper_problem1_reference.py to generate the Paper Problem 1 upstream reference via a batch-safe MATLAB wrapper, without editing the upstream clone.
• Extend upstream .mat loading to preserve full-reference fields (policy, tsw) and expose metrics needed for initialization/comparison.
• Add compare_paper_problem1_trajectories() for Pyomo-vs-upstream drying time, first switch time, terminal interface position, and max-temperature deviations.
• Document the generation and comparison commands in the Paper OCP docs and Pyomo models README.

Tests run

• python -m ruff check lyopronto/pyomo_models/paper_ocp.py lyopronto/pyomo_models/__init__.py benchmarks/paper_problem1_reference.py tests/test_pyomo_models/test_paper_ocp.py tests/test_pyomo_models/test_init.py - passed.
• pytest tests/test_pyomo_models/test_paper_ocp.py tests/test_pyomo_models/test_init.py -k 'load_upstream or compare_paper_problem1 or reference_runner or pyomo_exports' -n 0 - 5 passed, 14 deselected.
• pytest tests/test_pyomo_models/test_paper_ocp.py -m 'not slow' -n 0 - 11 passed, 3 deselected.
• pytest tests/test_pyomo_models/test_init.py -n 0 - 5 passed.
• python benchmarks/paper_problem1_reference.py generate --runner-only --work-dir /tmp/lyopronto-paper-problem1-runner --upstream-root /home/bernalde/repos/simDAE-optimalcontrol-lyo --output /tmp/lyopronto-paper-problem1-reference.mat - passed; wrote the batch-safe MATLAB wrappers and printed the matlab -batch command.
• git diff --check and git diff --cached --check - passed.

Notes about tests not run

• Full MATLAB/GEKKO reference generation was not run locally because there is no Linux matlab executable on PATH; the runner-only mode verified generation of the MATLAB wrapper files and command.
• pytest tests/test_pyomo_models -m 'not slow' -n 0 --maxfail=1 failed during collection with ModuleNotFoundError: No module named 'tests.utils'; direct Python imports can resolve tests.utils.
• Retried broader scope with python -m pytest tests/test_pyomo_models -m 'not slow' -n 0 --maxfail=1; it progressed through existing optimizer replay tests and was stopped after it became impractically long for this issue pass.

Closes #27