Fix #765: Upgrade 17 decision models to optimization (Group A + B)

[isPANN]

Summary

• Upgrade all 17 decision models from Value = Or + threshold to true optimization (Value = Max<T> / Min<T>)
• Rewrite 7 ILP rules from feasibility to optimization (add objective function, remove threshold constraint)
• Redesign config space for 3 models with variable-length outputs (padding encoding)
• Update CLI create commands, canonical examples, paper entries, and all tests

Group A — single-threshold models (12)

Model	Value	ILP	Config redesign
LongestCircuit	Max<W::Sum>
OptimalLinearArrangement	Min<usize>
RuralPostman	Min<W::Sum>
MixedChinesePostman	Min<W::Sum>
StackerCrane	Min<i32>
MinimumCardinalityKey	Min<i64>
SequencingToMinimizeMaximumCumulativeCost	Min<i64>
MultipleCopyFileAllocation	Min<i64>	✓
ExpectedRetrievalCost	Min<f64>	✓
SumOfSquaresPartition	Min<i64>	✓
LongestCommonSubsequence	Max<usize>	✓	✓ padding
ShortestCommonSupersequence	Min<usize>		✓ padding

Group B — constrained-optimization models (5)

Model	Keep (constraint)	Optimize	Value	ILP
MinimumCutIntoBoundedSets	size_bound	cut weight	Min<W::Sum>
ShortestWeightConstrainedPath	weight_bound	path length	Min<N::Sum>	✓
CapacityAssignment	delay_budget	cost	Min<u128>	✓
MinMaxMulticenter	k (centers)	max distance	Min<W::Sum>	✓ minimax
LengthBoundedDisjointPaths	max_length	# paths	Max<usize>

Test plan

• make check (fmt + clippy + test) passes
• cargo test --lib --features example-db — 3206 passed
• cargo test -p problemreductions-cli — 17 passed
• cargo test --test main (integration + doc tests) — all passed
• cargo clippy --all-targets — 0 warnings

Closes #765

🤖 Generated with Claude Code

[codecov]

Codecov Report

❌ Patch coverage is 99.30716% with 9 lines in your changes missing coverage. Please review.
✅ Project coverage is 97.82%. Comparing base (88fb8f7) to head (b70a7fd).

Files with missing lines	Patch %	Lines
src/models/graph/rural_postman.rs	75.00%	4 Missing ⚠️
...c/models/graph/shortest_weight_constrained_path.rs	88.23%	2 Missing ⚠️
src/models/graph/mixed_chinese_postman.rs	94.11%	1 Missing ⚠️
src/models/misc/longest_common_subsequence.rs	96.87%	1 Missing ⚠️
...t_tests/rules/shortestweightconstrainedpath_ilp.rs	85.71%	1 Missing ⚠️

Additional details and impacted files

@@           Coverage Diff            @@
##             main     #771    +/-   ##
========================================
  Coverage   97.82%   97.82%            
========================================
  Files         588      588            
  Lines       66422    66198   -224     
========================================
- Hits        64976    64760   -216     
+ Misses       1446     1438     -8     

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.

[isPANN]

Finding: ILP solver hangs due to unbounded variable domains

During this PR, we discovered that ILP<i32> reductions can cause HiGHS to hang indefinitely, even on tiny instances (e.g., 4 vertices / 23 variables).

Root cause

The ILP solver sets variable upper bounds from V::DIMS_PER_VAR:

v = v.max((V::DIMS_PER_VAR - 1) as f64);  // i32 → 2,147,483,647

For ILP<i32>, every variable gets domain [0, 2^31]. Even though constraints like x_{v,p} <= 1 or p_v <= n-1 logically bound variables to small ranges, HiGHS receives the full [0, 2^31] domain and its branch-and-bound becomes extremely slow.

This particularly affects auxiliary variables that lack explicit single-variable <= C constraints:

• z variables in minimax formulations (OLA z_e, Sequencing z)
• position/order variables only bounded via linking constraints (p_v = Σ p·x_{v,p})

Fix (3 parts)

1. ILP solver (src/solvers/ilp/solver.rs): Before passing to HiGHS, scan constraints for single-variable x_i <= C patterns and use the tighter bound instead of DIMS_PER_VAR
2. OLA ILP: Added z_e <= n-1 (max position difference)
3. Sequencing ILP: Added z <= Σ|costs| (max cumulative cost)

Impact

Before fix: model_specs_are_optimal hangs (OLA canonical example triggers OLA→ILP reduction path, HiGHS never returns).
After fix: all example-db tests pass in ~2 seconds.

Recommendation for future ILP rules

Always add explicit var <= upper_bound constraints for non-binary variables. This ensures HiGHS gets tight bounds regardless of DIMS_PER_VAR. See #772 for remaining hardcoded examples that should be converted.