Raise extras OptimizationOptimJL floor to 0.4.13

[ChrisRackauckas-Claude]

Please ignore this PR until it has been reviewed by @ChrisRackauckas.

Raise extras OptimizationOptimJL floor from 0.4.12 to 0.4.13.

#147 merged 0.4.12, but registered 0.4.12 (tree rVgAW) still does opt_ret = Symbol(Optim.converged(...)). Current SciMLBase OptimizationSolution requires ReturnCode.T. First registered _optim_retcode is 0.4.13 (tree mRLFy).

OptimizationOptimisers 0.3.14 still resolves and loads (the @SciMLMessage hole is 0.3.17–0.3.20). ModelingToolkit 11.7.1, ModelingToolkitBase 1.31.0, and SciMLSensitivity 7.93.0 also still resolve, so they are not changed.

Fail before

Isolated depot, Julia 1.10.11, Pkg.develop ModelingToolkitNeuralNets + pin OptimizationOptimJL 0.4.12, then solve(..., NelderMead()):

RESOLVE_OK PKG=OptimizationOptimJL VER=0.4.12
LOAD_OK MOD=OptimizationOptimJL
USE_FAIL
MethodError: Cannot `convert` an object of type Symbol to an object of type SciMLBase.ReturnCode.T

Pass after

Same graph + pin 0.4.13:

RESOLVE_OK PKG=OptimizationOptimJL VER=0.4.13
LOAD_OK MOD=OptimizationOptimJL
USE_OK RET=MaxIters TYPE=SciMLBase.ReturnCode.T

First working registered version is 0.4.13, not 0.4.14.