Solve documentation

[ocots]

We could add examples with the most common options. Or a list: print_level, tolerances...

[ocots]

@PierreMartinon Feel free to update the tutorial or to list here the most common options you think we can present / list.

[ocots]

Here or elsewhere, we should add a word about:

• selecting the linear solver
• use of HSL.jl
• use of MKL.jl

[PierreMartinon]

Most common options are now described in the CTDirect API doc and the solve-tutorial.
More technical options such as linear solver choice / HSL / MKL are not presented yet.

[ocots]

@PierreMartinon We can

• add the docstrings of the solve method,
• add options examples,
•  give a list of classical options: tol, max_iter
•  explain how to select the linear solver.

[ocots]

@jbcaillau Add ExaModels usage, GPU...

[ocots]

I have started to explain in the solve manual that ExaModels can be used: see #577.

We need to explain that the dynamics must be componentwise: maybe here.

We can give a word about GPU.

[jbcaillau]

hi @ocots i do not see your modification to the doc. i've created a branch, can you please merge what you've done in there? #580

[ocots]

I just wrote this

OptimalControl.jl/docs/src/manual-solve.md

Line 84 in 0497d90

- `:exa`: the NLP problem is modeled by a [`ExaModels.ExaModel`](@extref). It provides automatic differentiation and [SIMD](https://en.wikipedia.org/wiki/Single_instruction,_multiple_data) abstraction.

and this

OptimalControl.jl/docs/src/manual-solve.md

Lines 90 to 112 in 0497d90

	!!! warning
	The dynamics must be defined coordinatewise to use ExaModels.jl (`:exa`).
	For instance, let us try MadNLP solver with ExaModel modeller.
	```@example main
	using MadNLP
	ocp = @def begin
	t ∈ [ t0, tf ], time
	x = (q, v) ∈ R², state
	u ∈ R, control
	x(t0) == x0
	x(tf) == [0, 0]
	∂(q)(t) == v(t)
	∂(v)(t) == u(t)
	0.5∫( u(t)^2 ) → min
	end
	solve(ocp, :exa, :madnlp)
	nothing # hide
	```