Add benchmark validation and timeout guards

[bernalde]

Summary

• add post-solve benchmark metrics for final dryness shortfall, product-temperature violation, and optional Tsh/Pch ramp-rate violations
• mark benchmark records failed when those post-solve metrics violate configured constraints
• add benchmark CLI solver guards for IPOPT max_cpu_time and max_wall_time, propagate them through Pyomo adapters, and record configured timeout metadata
• document the benchmark CLI and JSONL schema changes
• filter known expected edge-case test warnings narrowly so the CI-equivalent test run is warning-clean

Closes #15
Closes #16

Tests run

• python -m compileall benchmarks lyopronto/pyomo_models tests/test_benchmarks.py - passed
• PYTHONPATH=. pytest tests/test_benchmarks.py -q - 12 passed
• PYTHONPATH=. pytest tests/ -m "not notebook and not pyomo" -q - 254 passed in 417.56s, no warning summary
• Warning triage run: PYTHONPATH=. pytest tests/ -m "not notebook and not pyomo" -q -o addopts='-n auto --tb=short --strict-markers --maxfail=1 --dist loadgroup' -W default - 254 passed, confirmed the warnings were expected legacy edge-case warnings

Notes

• pytest tests/test_benchmarks.py -q was also tried first, but this branch's pytest configuration does not put the repository root on PYTHONPATH; collection failed with ModuleNotFoundError: No module named 'benchmarks'. The tests were rerun with PYTHONPATH=..
• Follow-up issue for the broader legacy warning semantics: Audit legacy edge-case warning semantics #23.

[bernalde]

I found blocking issues in the benchmark validation behavior. GitHub does not allow this account to request changes on its own PR, but I would treat this as a request-changes review. The local targeted and CI-equivalent non-notebook/non-Pyomo tests pass, but I would not merge this until the blocking issues above are addressed.

[bernalde]

Pushed commit 53e18d2 (Address benchmark validation review feedback (#15, #16)).

Main changes made:

• Split benchmark metric kwargs so SciPy records keep product-temperature/dryness validation only, while Pyomo records receive the Pyomo-only ramp validation limits.
• Raised post-solve ramp validation tolerance to 1e-6 to align with solver feasibility tolerance and avoid false failures from accepted IPOPT numerical noise.
• Added focused benchmark tests for SciPy ramp-limit isolation and ramp noise within tolerance.

Tests run and results:

• PYTHONPATH=. pytest tests/test_benchmarks.py -q -> 14 passed.
• python -m compileall benchmarks lyopronto/pyomo_models tests/test_benchmarks.py -> passed.
• PYTHONPATH=. pytest tests/ -n auto -v -m "not notebook and not pyomo" --ignore=tests/test_pyomo_models -> 251 passed in 375.92s.

Comments intentionally not addressed:

• The nonblocking pytest.ini warning-filter concern remains intentionally unchanged in this PR. It is broader than benchmark validation/timeouts and is already tracked by Audit legacy edge-case warning semantics #23, including the follow-up to replace broad filters with local warning assertions where practical.

Remaining risks or follow-up items:

• Audit legacy edge-case warning semantics #23 still needs the deeper warning semantics audit and localized warning cleanup.