Dipole kick held in KickAngle is not seen by the magnet accessors #471

What happens

AT keeps the dipole kick of an element in two places: the polynom (PolynomB[0] / PolynomA[0]) and KickAngle. The magnet accessors in pyaml/lattice/abstract_impl.py read and write the polynom only, so a kick that lives in KickAngle is not seen:

The reason is in AT: CorrectorPass integrates KickAngle and ignores the polynoms; the multipole pass methods (ThinMPolePass, StrMPole*, BndMPole*, ExactMultipole*) integrate the polynom and add KickAngle to it.

#435 fixed the length bookkeeping for zero-length elements; this is a separate point. The thin-corrector test added there uses at.Corrector, so it checks that set and get agree but not that the orbit moves. I saw that #467 moved the TL2 model from CorrectorPass to ThinMPolePass, which avoids the first row for that lattice.

How to reproduce

import at, numpy as np
from pyaml.lattice.abstract_impl import RWStrengthScalar
from pyaml.magnet.hcorrector import HCorrector
from pyaml.magnet.identity_model import IdentityMagnetModel

def ring(cor):
    qf, qd = at.Quadrupole("QF", 0.5, 1.2), at.Quadrupole("QD", 0.5, -1.2)
    return at.Lattice([cor] + [qf, at.Drift("D", 1.0), at.Monitor("BPM"), qd, at.Drift("D", 1.0)] * 8,
                      energy=2e9, periodicity=1)

def orbit(cor):
    return np.abs(at.find_orbit4(ring(cor), refpts=at.Monitor)[1][:, 0]).max()

model = IdentityMagnetModel(physics="COR", unit="rad")
for cor in (at.Corrector("C", 0.0, [0.0, 0.0], PolynomA=[0.0], PolynomB=[0.0]),
            at.ThinMultipole("C", [0.0], [0.0], KickAngle=[1e-4, 0.0]),
            at.ThinMultipole("C", [0.0], [0.0])):
    acc = RWStrengthScalar([cor], HCorrector.polynom, model)
    before = (acc.get(), orbit(cor))
    acc.set(1e-4)
    print(f"{cor.PassMethod}: before get={before[0]:g} orbit={before[1]:.3e} | after get={acc.get():g} orbit={orbit(cor):.3e}")

Output on main (AT 0.8.0); 1.518e-3 m is the orbit of a 1e-4 rad kick in this ring:

CorrectorPass: before get=0 orbit=0.000e+00 | after get=0.0001 orbit=0.000e+00
ThinMPolePass: before get=0 orbit=1.518e-03 | after get=0.0001 orbit=3.036e-03
ThinMPolePass: before get=0 orbit=0.000e+00 | after get=0.0001 orbit=1.518e-03

Possible directions

1. Support it: the accessors address the dipole component through KickAngle where the pass method uses it (write KickAngle for CorrectorPass; read polynom + KickAngle for the multipole passes and reset the KickAngle component on write).
2. Refuse it: reject such elements when the simulator is loaded, with a message that says to model correctors as (thin) multipoles without a KickAngle.

Either way, an orbit-based test would keep it from coming back.

I opened #470 with an implementation of direction 1 before reading the contribution guide — apologies for the order. I am happy to change it to direction 2, or to whatever you prefer after discussion here.