[Dev] Solve update

[jbcaillau]

@ocots @PierreMartinon a few updates needed (provided i tried and used correctly solve with the new flat syntax):

• the error is not as nice as before when the user forgets using NLPModelsIpopt
  
• same with(out) MadNLP
  
• looks like we have to both using MadNLP and using MadNLPMumps to trigger the extension... (anyway this has to be updated with MadNLP new release v0.9)
  
• similar issue with MadNLPGPU; plus it tries to use Mumps on CUDA! with backend=CUDABackend(), it only makes sense to use linear_solver=MadNLPGPU.CUDSSSolver which should then be the default in that case
  

[jbcaillau]

@ocots subissue no. 1 ✅

julia> solve(o)
ERROR: Control Toolbox Error

❌ Error: CTBase.Exceptions.ExtensionError, missing dependencies to access Ipopt{CTSolvers.Strategies.CPU} options metadata
📦 Missing dependencies: NLPModelsIpopt
💡 Suggestion: julia> using NLPModelsIpopt

[jbcaillau]

@ocots subissue(s) no. 2 ✅

julia> solve(o, :madnlp)
ERROR: Control Toolbox Error

❌ Error: CTBase.Exceptions.ExtensionError, missing dependencies to access MadNLP{CTSolvers.Strategies.CPU} options metadata
📦 Missing dependencies: MadNLP
💡 Suggestion: julia> using MadNLP

julia> using MadNLP

julia> solve(o, :madnlp)
▫ OptimalControl v1.2.3-beta solving with: collocation → adnlp → madnlp (cpu)

  📦 Configuration:
   ├─ Discretizer: collocation (no user options)
   ├─ Modeler: adnlp (no user options)
   └─ Solver: madnlp (linear_solver = MumpsSolver [cpu-dependent])

julia> solve(o, :gpu)
▫ OptimalControl v1.2.3-beta solving with: collocation → exa → madnlp (gpu)

  📦 Configuration:
   ├─ Discretizer: collocation (no user options)
   ├─ Modeler: exa (backend = CUDABackend(false, false) [gpu-dependent])
   └─ Solver: madnlp (linear_solver = CUDSSSolver [gpu-dependent])

julia> isdefined(Main, :CUDA)
false

julia> using CUDA

julia> isdefined(Main, :CUDA)
true

julia> solve(o, :gpu)
ERROR: Control Toolbox Error

❌ Error: CTBase.Exceptions.ExtensionError, missing dependencies to use GPU linear solver with MadNLP/MadNCL
📦 Missing dependencies: MadNLPGPU
💡 Suggestion: julia> using MadNLPGPU

julia> using CUDA

julia> solve(o, :gpu)
▫ OptimalControl v1.2.3-beta solving with: collocation → exa → madnlp (gpu)

  📦 Configuration:
   ├─ Discretizer: collocation (no user options)
   ├─ Modeler: exa (backend = CUDABackend(false, false) [gpu-dependent])
   └─ Solver: madnlp (linear_solver = CUDSSSolver [gpu-dependent]

[jbcaillau]

@ocots still a minor issue on these cases (should indicate to choose the non-default combo :exa for the modeller and :madnlp for the solver, which is the only one that allows to use a CUDA backend):

julia> using MadNLPGPU

julia> solve(o; backend=CUDABackend())
┌ Error: Validation failed for option backend with value CUDABackend(false, false)
│   exception =
│    MethodError: no method matching (::CTSolvers.Modelers.var"#get_validate_adnlp_backend##0#get_validate_adnlp_backend##1"{Type{CTSolvers.Modelers.ADNLPTag}})(::CUDABackend)
│    The function `#get_validate_adnlp_backend##0` exists, but no method is defined for this combination of argument types.
│    
│    Closest candidates are:
│      (::CTSolvers.Modelers.var"#get_validate_adnlp_backend##0#get_validate_adnlp_backend##1")(::Symbol)
│       @ CTSolvers ~/.julia/packages/CTSolvers/Sfdxv/src/Modelers/adnlp.jl:552

julia> solve(o, :exa; backend=CUDABackend())
┌ Warning: Inconsistent backend (CUDABackend) for CPU parameter
└ @ CTSolvers.Modelers ~/.julia/packages/CTSolvers/Sfdxv/src/Modelers/exa.jl:259
▫ OptimalControl v1.2.3-beta solving with: collocation → exa → ipopt (cpu)

  📦 Configuration:
   ├─ Discretizer: collocation (no user options)
   ├─ Modeler: exa (backend = CUDABackend(false, false))
   └─ Solver: ipopt (no user options)

▫ ERROR: MethodError: no method matching CreateIpoptProblem(::Int64, ::CuArray{…}, ::CuArray{…}, ::Int64, ::CuArray{…}, ::CuArray{…}, ::Int64, ::Int64, ::NLPModelsIpopt.var"#eval_f#set_callbacks##0"{…}, ::NLPModelsIpopt.var"#eval_g#set_callbacks##1"{…}, ::NLPModelsIpopt.var"#eval_grad_f#set_callbacks##2"{…}, ::NLPModelsIpopt.var"#eval_jac_g#set_callbacks##3"{…}, ::NLPModelsIpopt.var"#eval_h#set_callbacks##4"{…})
The function `CreateIpoptProblem` exists, but no method is defined for this combination of argument types.