Inconsistent type enforcement for finalize argument in ground state solvers

[20akshay00]

DMRG methods take in a keyword argument finalize, while GradientGrassmann take in finalize!, meant to act as a callback function at the end of each solver iteration. DMRG is flexible with the type of the argument, whereas GradientGrassmann casts it into a Function.

This was encountered when trying to use a struct as a function so that it carries local state while performing the required finalizations.

[lkdvos]

Additionally, the signatures are also very different:
OptimKit uses finalize!(x, f, g, numiter) -> x, f, g, while DMRG requires finalize(iter, state, H, envs) -> state, envs

[20akshay00]

Yes, I noticed that also, but thats just by virtue of DMRG working differently than Gradient methods whereas the casting as far as I see, isn't strictly necessary (?).

However, I did wonder if there was a scope to introduce a unified callback interface for these solvers, like how DifferentialEquations.jl does it. But I will open a separate issue for that.