Reorganize amplitude building functions and make public

[redeboer]

For the documentation (but arguably also for the design), it would be more transparent to build the amplitude model through functions. In particular, the _generate_partial_decay methods can be reformulated so that they just produce a product of terms that are relevant to the helicity kinematics only (no dynamics).

This issue was motivated by ComPWA/qrules#57, where it would have been nice to explain how the InteractionProperties are used to create Clebsch-Gordan coefficients in the canonical basis by pointing to the functions responsible for this in AmpForm:

ampform/src/ampform/helicity.py

Lines 584 to 589 in cdec30b

	def _generate_partial_decay( # pylint: disable=too-many-locals
	self, graph: StateTransitionGraph[ParticleWithSpin], node_id: int
	) -> sp.Expr:
	wigner_d = self._generate_wigner_d(graph, node_id)
	dynamics_symbol = self.__create_dynamics(graph, node_id)
	return wigner_d * dynamics_symbol

ampform/src/ampform/helicity.py

Lines 671 to 673 in cdec30b

	def _generate_partial_decay( # pylint: disable=too-many-locals
	self, graph: StateTransitionGraph[ParticleWithSpin], node_id: int
	) -> sp.Symbol:

[redeboer]

This may also make it easier to write some unit tests for the Clebsch-Gordan coefficients. These arguments j1 through to m3 are rather error-prone:

ampform/src/ampform/helicity.py

Lines 707 to 723 in cdec30b

	cg_ls = CG(
	j1=sp.nsimplify(ang_mom.magnitude),
	m1=sp.nsimplify(ang_mom.projection),
	j2=sp.nsimplify(spin.magnitude),
	m2=sp.nsimplify(decay_particle_lambda),
	j3=sp.nsimplify(parent_spin.magnitude),
	m3=sp.nsimplify(decay_particle_lambda),
	)
	cg_ss = CG(
	j1=sp.nsimplify(daughter_spins[0].magnitude),
	m1=sp.nsimplify(daughter_spins[0].projection),
	j2=sp.nsimplify(daughter_spins[1].magnitude),
	m2=sp.nsimplify(-daughter_spins[1].projection),
	j3=sp.nsimplify(spin.magnitude),
	m3=sp.nsimplify(decay_particle_lambda),
	)
	return cg_ls * cg_ss * amplitude

[redeboer]

Largely addressed through #87, which introduced formulate_clebsch_gordan_coefficients() and formulate_wigner_d(). With #245, however, the HelicityAmplitudeBuilder has again acquired a lot of implementation methods that are note well documented.