Compare POUNCE 0.10.0 across NLP applications

[Mazhar331]

Summary

• add a pounce extra pinned to pounce-solver==0.10.0 and pyomo-pounce==0.10.0
• add an automatic, path-scoped POUNCE solver lane for non-draft Pyomo-sensitive changes, while retaining local and manual-dispatch entry points
• exercise the complete optional Pyomo application suite, including solver-backed notebooks, through the same AMPL/ASL interface used for IPOPT
• record and test the POUNCE pseudosteady-limit baseline beside IPOPT
• extend solver-backed coverage to the remaining parameter-estimation, design-space, and multi-vial applications
• document solver-status differences and keep the known Problem 2 GDP restart instability visible as a strict xfail

The POUNCE and IPOPT jobs remain separate CI results. Both use the existing Pyomo scope decision, run for relevant non-draft changes, and report a clean skip elsewhere so either can later become a required check. POUNCE does not replace IPOPT or become a default package dependency.

Results

Across the shared successful pseudosteady rungs, POUNCE and IPOPT agree within the committed scientific tolerances. The application-wide lane passes except for one deliberate strict xfail:

• POUNCE #590: released 0.10.0 reports the feasible Problem 1 f=0.02 rung as infeasible. The proposed upstream fix in #596 has been validated downstream, but is not released.
• POUNCE #592: one Problem 2 fixed-policy NLP reports success before its solution is restart-stable. The exact GPL-licensed .nl reproducer is available upstream.

This PR intentionally pins the released 0.10.0 behavior. A later POUNCE release should be handled in a focused follow-up that regenerates the baseline and removes the xfail only when the original cold GDP path is stable.

Validation

• actionlint .github/workflows/pyomo-tests.yml .github/workflows/slow-tests.yml
• python -m ruff check lyopronto tests examples main.py
• pytest -n 0 -q tests/test_packaging_config.py tests/test_pounce_solver_comparison.py — 21 passed
• local SKIP_INSTALL=1 ./run_local_ci.sh pounce — 183 passed, 1 xfailed, 418 deselected
• exact-head GitHub Actions run — POUNCE 0.10.0 installed and the POUNCE solver lane passed with 183 passed, 1 xfailed, 418 deselected

Stacking

This PR is intentionally based on feat/issue-146-surface-convergence-quality / #147 because the comparison consumes the convergence-quality schema introduced there. Retarget it to main after #147 lands.