Feature/merge trajectories

[andgoldschmidt]

The purpose of this PR is to eliminate broadcasting in trajectory initialization by using the NamedTrajectories merge call, see kestrelquantum/NamedTrajectories.jl#54

The price to pay is some copying.

The benefit is that this puts the burden of initialization for problems like direct sums and sampling onto the problem template. It forces the problem template to be aware of naming, which is good for programming at the level of problem templates.

For example, a clean loop for setting up a sampling problem with a lot of complexity like separate goals, separate systems, and so on:

    # Trajectory
    state_names = [
        Symbol(string(state_name, "_system_", label)) for label ∈ 1:length(systems)
    ]

    if !isnothing(init_trajectory)
        traj = init_trajectory
    else
        trajs = map(zip(systems, operators, state_names)) do (sys, op, s)
            initialize_trajectory(
                op,
                T,
                Δt,
                length(sys.G_drives),
                (a_bounds, da_bounds, dda_bounds);
                state_name=s,
                control_name=control_name,
                timestep_name=timestep_name,
                free_time=piccolo_options.free_time,
                Δt_bounds=(Δt_min, Δt_max),
                geodesic=piccolo_options.geodesic,
                bound_state=piccolo_options.bound_state,
                a_guess=a_guess,
                system=sys,
                rollout_integrator=piccolo_options.rollout_integrator,
            )
        end

        traj = merge(
            trajs, 
            merge_names=(; a=1, da=1, dda=1, Δt=1),
            free_time=piccolo_options.free_time
        )
    end    

    control_names = [
        name for name ∈ traj.names
        if endswith(string(name), string(control_name))
    ]