[Dev] Remove the solver from `direct_transcription` description

[ocots]

Do not pass the nlp_solver to direct_transcription:

CTDirect.jl/src/solve.jl

Lines 244 to 254 in 6e0be4d

	function direct_transcription(
	ocp::CTModels.Model,
	description...;
	grid_size=__grid_size(),
	disc_method=__disc_method(),
	time_grid=__time_grid(),
	init=__ocp_init(),
	lagrange_to_mayer=true,
	kwargs...,
	)
	nlp_solver, nlp_model = parse_description(description)

Need to split the parse_description to get separately the solver and the model:

CTDirect.jl/src/solve.jl

Lines 80 to 83 in 6e0be4d

	function parse_description(description)
	# default: Ipopt, ADNLPModels
	method = CTBase.complete(description; descriptions=available_methods())

The line:

method = CTBase.complete(description; descriptions=available_methods())

is useful to check if the description is consistent with available_methods() but the list of available methods is for the solve method.

Do we create a available_methods_for_discretisation()?

[PierreMartinon]

Hi Olivier,

Done in #487, the nlp_solver was a remnant of an old test with AD backends for matrix-free solvers such as Percival (disappointing, see Percival discussion), and is not needed for the usual usage.

Method parser calls now look like

nlp_solver = parse_description(description, :solver)
nlp_model = parse_description(description, :model)

About the last part, not sure there is a point in having different sets of methods ?

[ocots]

Not sure to understand your last question.

[PierreMartinon]

Not sure to understand your last question.

:D I was referring to available_methods_for_discretization. Not sure it is worth it to introduce another set of methods. The unused information will just be ignored anyway, unless I missed something ?