Replace array coefficients with Python function and (in the future) qutip.Coefficient

[BoxiLi]

Currently, the coefficient of a time-dependent Hamiltonian is represented by a NumPy array and interpolation to support arbitrary non-analytical functions. However, in some circumstances, a Python function might be preferable. Hence, it would be good to replace the internal representation of coefficients with a more general python function representation. This will still be compatible with the old input API by converting the NumPy array to a SciPy spline function.

• In some circumstances, a Python function might be preferable. E.g. if the control coefficients are analytically defined and need to be very accurately calculated, the number of samples in the spline interpolation needs to be very high, which increases the memory cost. This is also relevant for simulating gate operations with very high fidelity.

• Also, in qutip 5, the newly introduced Coefficient class makes it more flexible to define different kinds of coefficients. A Coefficient instance behaves exactly like a Python function, with a __call__ method. Hence, making this update will also make it easier for qutip-qip to use the new Coefficient interface of qutip v5.

[BoxiLi]

I tested it very quickly by replacing all the NumPy array coefficients with SciPy interp1d when generating QobjEvo. It works, however, the tests become 5~10 times slower... I suspect this is because QobjEvo needs to switch between Python and Cython. The effect is probably exaggerated by the small size of the tests examples. But still this seems to be quite a large compromise on performance.