support `optimize` in builder

[Roger-luo]

most of the optimization based problem

# f(report)
job.optimize(f, CMAES()).submit().report()

this should cover most of the optimization problem we have on pulses

[weinbe58]

I think we need to refactor the builder a bit to create an intermediate object that can be used to dynamically generate tasks/jobs and forward this from Emit. I think the following would be idea: the objects for generating jobs are constructed by methods from Emit, the argument of this method always being the static_assignments which correspond to the arguments of assign later the batch_assign arguments are passed in during the submit() method of this Job generator object.

The motivation for this is because in optimize I need to have some object that can generate tasks on the fly and I don't want to re-implement the same infrastructure.

job = (
    Square(4, lattice_spacing="a")
    .apply_defect_density(0.1)
    .rydberg.detuning.uniform.piecewise_linear(
        durations=[0.1, 3.8, 0.1], values=[-10, -10, "final_detuning", "final_detuning"]
    )
    .rabi.amplitude.uniform.piecewise_linear(
        durations=[0.1, 3.8, 0.1], values=[0.0, 15.0, 15.0, 0.0]
    )
    .braket(**static_assignments)
    .submit(100, **batch_assignments)
)

# optimization loop

optimized_results = (
    Square(4, lattice_spacing="a")
    .apply_defect_density(0.1)
    .rydberg.detuning.uniform.piecewise_linear(
        durations=[0.1, 3.8, 0.1], values=[-10, -10, "final_detuning", "final_detuning"]
    )
    .rabi.amplitude.uniform.piecewise_linear(
        durations=[0.1, 3.8, 0.1], values=[0.0, 15.0, 15.0, 0.0]
    )
    .braket(**static_assignments)
    .optimize(100, cost_function)
)

[weinbe58]

Oh one issue would be on how to deal with RunTimeVectors in the optimization step.