CompatHelper: bump compat for NLPModels to 0.22 for package OptimizationCUTEst, (keep existing compat)

[github-actions]

This pull request changes the compat entry for the NLPModels package from 0.21 to 0.21, 0.22 for package OptimizationCUTEst.
This keeps the compat entries for earlier versions.

Note: I have not tested your package with this new compat entry.
It is your responsibility to make sure that your package tests pass before you merge this pull request.

[ChrisRackauckas-Claude]

Analysis of CI Failure

The CI build failed after running for 6+ days with exit status -1 (timeout/termination), not due to a specific code error from NLPModels API changes.

NLPModels 0.22 Breaking Changes

The release notes indicate these breaking changes:

1. Dropped Julia 1.6 support (now requires Julia 1.10+)
2. jac_lin* functions no longer take an x argument
3. objcons! with in-place residual not defined for NLS

Impact Assessment

The benchmark code does NOT directly use any of the affected APIs:

• Uses CUTEstModel from CUTEst.jl (wrapper)
• Uses OptimizationNLPModels.OptimizationProblem
• Only accesses nlp_prob.meta.nvar and calls finalize(nlp_prob)
• No direct calls to jac_lin* or objcons!

Root Cause

The CI timeout appears to be due to the OptimizationCUTEst benchmark suite being resource-intensive (runs many optimization problems with multiple optimizers over several days), rather than any breaking changes from NLPModels 0.22.

Recommendation

This compat bump should be safe to merge. The benchmark timeout is a pre-existing infrastructure concern, not a compatibility issue. The downstream packages (CUTEst.jl, OptimizationNLPModels.jl) handle the NLPModels interface internally and would need to be updated if there were real compatibility issues.

cc @ChrisRackauckas

🤖 Generated with Claude Code

[ChrisRackauckas-Claude]

BuildKite Failure Analysis

This PR attempted to bump NLPModels from 0.21 to 0.22 for the OptimizationCUTEst benchmarks.

Status: Failed in BuildKite build #3864 (November 27, 2025)

What Failed:

• The benchmark job :hammer: benchmarks/OptimizationCUTEst crashed with exit status -1
• The job ran on AMD CI server but never completed successfully
• Standard test suite passed, only benchmark execution failed

Root Cause:
The NLPModels v0.22 update appears to have introduced incompatibilities that prevent the OptimizationCUTEst benchmark suite from running successfully.

Recommendation:
This PR should remain on hold until:

1. The incompatibility between NLPModels v0.22 and OptimizationCUTEst benchmarks is investigated
2. Either the benchmarks are updated to work with NLPModels v0.22, or
3. The NLPModels dependency is confirmed to need to stay at v0.21

cc @ChrisRackauckas