Refactor spin-flip TDDFT

[wtpeter]

This PR refactors the SF-TDDFT and SF-TDA modules to improve compatibility, functionality, and code structure.

Changes include:

1. Fixed an incompatibility issue regarding the max_space parameter with recent PySCF versions.
2. Added support for collinear functionals when collinear_samples < 0.
3. Refactored SF-TDA and SF-TDDFT code to remove redundancies. The implementation now aligns better with current PySCF and GPU4PySCF conventions.
4. Refined the pickeig function for SF-TDA and SF-TDDFT; the positive_eig_threshold parameter is no longer needed.
5. Added support for ROKS reference states in SF-TDA and SF-TDDFT excitation energy calculations (note: analytic gradients are not yet supported).
6. Relaxed test tolerances to 1e-4 to account for current instability in the diagonalization solver.
7. Fixed errors in the spin square (<S^2>) calculation and added utility features, including oscillator strength, symmetry analysis, and transition mode analysis.

NOTE: It addresses the compatibility issue mentioned in #173 but with more modifications.

[jeanwsr]

Thanks for the fix. For collinear functional, i think it may be easier to understand if using gpu4pyscf's definition: td.collinear = 'col'.

[wtpeter]

Thanks for the fix. For collinear functional, i think it may be easier to understand if using gpu4pyscf's definition: td.collinear = 'col'.

Thanks for the suggestion! That would indeed be more consistent with gpu4pyscf. For the current refactor, I would keep the collinear_samples < 0 logic for now, and I will refine the API definition in a follow-up update.

[jeanwsr]

Suggested change

	Generate a function to compute the product of Spin-flip UKS response function
	Generate a function to compute the product of Spin-flip UHF/UKS response function