Allow SymbolicAnalysis 0.5 in ConvexOptimization

[ChrisRackauckas-Claude]

Allow ConvexOptimization to use SymbolicAnalysis 0.5 while retaining 0.3 support. OptimizationBase requires 0.5, so developing both subpackages currently fails before ModelingToolkit downstream tests can run. This is a compatibility-only change; the existing analyze signature and documented result fields used here remain available in 0.5.1.

Fixes #1352.

Please ignore this PR until reviewed by @ChrisRackauckas.

Verification on Julia 1.12.7:

JULIA_PKG_PRECOMPILE_AUTO=0 julia --project=resolver-repro -e 'using Pkg; Pkg.develop([PackageSpec(path="lib/OptimizationBase"), PackageSpec(path="lib/ConvexOptimization")]); Pkg.status(); println("JOINT RESOLUTION PASSED"); ENV["OPTIMIZATION_TEST_GROUP"]="Core"; Pkg.test("ConvexOptimization")'

Before (clean master and introducing commit, fresh local environment):

ERROR: Unsatisfiable requirements detected for package SymbolicAnalysis [4297ee4d]:
 SymbolicAnalysis [4297ee4d] log:
 ├─possible versions are: 0.1.0 - 0.5.1 or uninstalled
 ├─restricted to versions 0.5 by OptimizationBase [bca83a33], leaving only versions: 0.5.0 - 0.5.1 or uninstalled
 └─restricted to versions 0.3 by ConvexOptimization [e7fa4f8d] — no versions left

After (same pair-development operation, fresh local environment, followed by the existing complete Core group):

[4297ee4d] SymbolicAnalysis v0.5.1
Test Summary: | Pass  Total     Time
Core          |  196    196  4m47.2s
Testing ConvexOptimization tests passed
JOINT RESOLUTION PASSED

typos lib/ConvexOptimization/Project.toml and git diff --check both exited 0. No Julia source changed, so Runic does not apply. This subpackage defines only Core and has no QA group. No documentation/public API changed, so no docs build was run. The optional pre-existing Convex.jl cross-check was unavailable; all analytic Core assertions ran. The full ModelingToolkit downstream suite, other Julia versions, and allowed-to-fail/GPU jobs were not run locally.

🤖 Generated with Codex CLI 0.153.4 (model: gpt-6-astra; local parent session ID: 01a08039-86d4-7683-9601-30fd79501e27).

[ChrisRackauckas-Claude]

CI completed for this compat-only commit: 50 successful checks, 1 skipped, 1 failed.

The ModelingToolkit downstream job now resolves SymbolicAnalysis 0.5.1 and passes. Its final log reports:

Test Summary: | Pass  Broken  Total       Time
Optimization  |  254       1    255  328m47.8s
Testing ModelingToolkit tests passed

The reported broken test is already in the downstream suite; this PR changes no tests. All package tests, documentation, and DiffEqFlux downstream checks also passed.

The sole failure is NeuralPDE NNPDE1, assertions at nnpde__pde_iv_system_of_pdes.jl:105,106 (0 passed, 2 failed), matching the separate master-failure investigation tracked at SciML/NeuralPDE.jl#1164. No assertion was loosened or skipped here.

Links:

• Passing ModelingToolkit downstream: https://github.com/SciML/Optimization.jl/actions/runs/34685366092/job/103531324575
• NeuralPDE failure: https://github.com/SciML/Optimization.jl/actions/runs/34685366092/job/103531324591
• NeuralPDE investigation: NNPDE1 PDE IV test is underdetermined on the full square (standalone counterexample) NeuralPDE.jl#1164

The PR remains a draft and should be ignored until reviewed by @ChrisRackauckas.

🤖 Recorded by Codex CLI 0.153.4 (model: gpt-6-astra; local parent session ID: 01a08039-86d4-7683-9601-30fd79501e27).