[Ongoing] Test for eta suitable for Sella-ABACUS

Sella is using finite difference method to generate initial Hessian for TS search (dimer likely), which is using a relatively small stepsize `eta=1e-4`, but this is not suitable for ABACUS and any other DFT software. For finite difference, a large stepsize will lead to the formula break-down and the calculation break-down, but a small stepsize will make the intrinsic error of Energy/Force calculation larger and the calculation will fail. It is needed to reach a balance for stepsize. It is needed to do some test for Sella stepsize. Especially for ABACUS usage. - For dimer, the stepsize is the length of dimer, which tend to be 1e-2 ~ 5e-3 - For vibration analysis, the stepsize tend to be 1e-2 - For differential test in ABACUS, the stepsize tend to range from 1e-2 to 1e-3, 2e-4 may be the lowest. So the Stepsize in Sella now is relatively small Reference: https://github.com/zadorlab/sella/wiki/Hyperparameters https://en.wikipedia.org/wiki/Numerical_differentiation https://github.com/deepmodeling/abacus-develop/issues/3645