Add test for dual_ftol_rel parameter

- Test that dual_ftol_rel can be passed without UndefVarError
- Test with both NLopt.Opt and direct algorithm interfaces
- Verify it works combined with other solver options

🤖 Generated with [Claude Code](https://claude.ai/code)

Co-Authored-By: Claude <[email protected]>

Author: ChrisRackauckas

lib/OptimizationNLopt/test/runtests.jl

@@ -100,6 +100,29 @@ using Test, Random
         @test sol.retcode == ReturnCode.MaxTime
     end
 
+    @testset "dual_ftol_rel parameter" begin
+        # Test that dual_ftol_rel parameter can be passed to NLopt without errors
+        # This parameter is specific to MMA/CCSA algorithms for dual optimization tolerance
+        x0_test = zeros(2)
+        optprob = OptimizationFunction(rosenbrock, Optimization.AutoZygote())
+        prob = OptimizationProblem(optprob, x0_test, _p)
+        
+        # Test with NLopt.Opt interface
+        opt = NLopt.Opt(:LD_MMA, 2)
+        # This should not throw an error - the PR fixed the UndefVarError
+        sol = solve(prob, opt, dual_ftol_rel = 1e-6, maxiters = 100)
+        @test sol.retcode ∈ [ReturnCode.Success, ReturnCode.MaxIters]
+        
+        # Test with direct algorithm interface
+        sol = solve(prob, NLopt.LD_MMA(), dual_ftol_rel = 1e-5, maxiters = 100)
+        @test sol.retcode ∈ [ReturnCode.Success, ReturnCode.MaxIters]
+        
+        # Verify it works with other solver options
+        sol = solve(prob, NLopt.LD_MMA(), dual_ftol_rel = 1e-4, ftol_rel = 1e-6, 
+                   xtol_rel = 1e-6, maxiters = 100)
+        @test sol.retcode ∈ [ReturnCode.Success, ReturnCode.MaxIters]
+    end
+
     @testset "constrained" begin
         Random.seed!(1)
         cons = (res, x, p) -> res .= [x[1]^2 + x[2]^2 - 1.0]