ENH: skip spin-projection expansion

[grayson-helmholz]

• Closes Generate transitions without spin projections #219

⚙️ Enhancements

• create_qn_problem_sets() takes a spin_projections argument. With spin_projections=False, the Cartesian expansion over the spin-projection combinations of the initial and final state is never generated, instead of being generated and then discarded. The result is the same as passing the collection through strip_spin_projections(), but it avoids building the combinations in the first place, which dominates the cost for reactions with several spin-carrying states. Combining it with merge_spin_projections=True is rejected with a ValueError, because there is nothing left to merge.
• create_initial_facts() and create_problem_sets() expose the same switch as expand_spin_projections. When it is off, create_initial_facts() returns a single InitialFacts in which every external state has its spin projection set to None.

📝 Documentation

• New usage page usage/qn-transitions, which walks through solving a reaction at the $J^{P(C)}$ level: creating projection-free problem sets, solving them with find_qn_transitions(), and reading the resulting QNTransitions. QNTransition is registered in the Sphinx configuration so that it renders as a type alias in the API pages.

Squash commit messages

* DOC: add notebook on quantum-number transitions