[github-actions]

This pull request changes the compat entry for the MadNLPGPU package from 0.10 to 0.10, 0.11 for package docs.
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.

[ocots]

Setting this PR to draft: for now it does not pass, because of compatibility problems in other packages (MadNLPGPU 0.11 needs CUDSS 1 and KernelAbstractions 0.10: blocked by ArrayInterface and Enzyme).

Why (checked with minimal resolutions, details in #315): the four bumps CUDSS 1, MadNLPGPU 0.11, MadNLP 0.11 and ExaModels 0.14 go together (CUDSS 1 ← MadNLPGPU 0.11 ← MadNLP 0.11, and ExaModels 0.14 cannot sit next to MadNLPGPU 0.10 because of KernelAbstractions 0.9/0.10). They resolve fine among themselves with CUDA 6, but two packages of the test environment block them:

1. ArrayInterface (needed by SciMLBase 3, DiffEqBase 7, OrdinaryDiffEqTsit5 2): its weak dependency CUDSS is bounded ≤ 0.8 in every release up to 7.30.2. Upstream: CompatHelper: bump compat for CUDSS in [weakdeps] to 1, (keep existing compat) JuliaArrays/ArrayInterface.jl#513 (CompatHelper PR, open, not merged).
2. Enzyme 0.13 requires LLVM 9, while KernelAbstractions 0.10.4 (needed by MadNLPGPU 0.11) requires LLVM 10. Upstream: Upgrade to LLVM.jl 10 EnzymeAD/Enzyme.jl#3741 "Upgrade to LLVM.jl 10" (open).

What unblocks it: a release of ArrayInterface with CUDSS 1 and a release of Enzyme with LLVM.jl 10. Then the triage is rerun and the bumps are tested, which may still show code changes.