[Dev] v1.1

[jbcaillau]

@ocots @PierreMartinon @gergaud some refactoring after adding new discretisation towards ExaModels (runs on GPU)

• keep track of the original expression (= Julia source code captured by @def) at all levels
• having the info allows enhanced actions (for solving, plotting...) - this is already what is done in CTModels, CTDirect, CTFlows
• parsing (syntactic + semantic) by @def creates an OptimalControl.Model that gives access to the standard getters (on state dim, dynamics, objective, etc.)
• essentially 4 actions : discretise, flow, solve, plot; some can be called at different steps (e.g. solve just after @def, or after discretise, or after flow...); see use cases below
• should be not breaking wrt. what is currently implemented in v1.0 (*)

(*) @PierreMartinon may be one suggested change: when discretising towards an nlp, rather than a model (ADNLP...) generate a function parametrised by grid_size, scheme, init... that, when evaluated, returns such a model (see use case no. 3 below; see also https://github.com/control-toolbox/CTParser.jl/blob/6bdf969c1140ea1497b305901218cd2eb7f6bfe6/src/onepass.jl#L1026) this allows in particular to have a PRAGMA in the generated (see naive example here) and is in line with what is often done in practice. to be discussed

# use case no.1

ocp = @def e
sol = solve(ocp)
plot(sol)

# use case no. 2

ocp = @def e
dx = dynamics(ocp)(t, x, u, v)
docp = discretise(ocp)
sol = solve(docp)
plot(sol)

# use case no. 3

ocp = @def e
docp = discretise(ocp; model = :exa)
nlp = docp(; grid_size = 200)
sol = madnlp(nlp)

# use case no. 4

ocp = @def e
f = flow(ocp, u)
sol = f((t0, tf), x0, p0)
plot(sol)