Integrate Krylov.jl for pressure solve

[xkykai]

This PR aims to integrate KrylovSolver for the pressure solve which has been developed into Oceananigans for the nonhydrostatic model. I begun by developing a new solver called KrylovPoissonSolver, which is a Poisson solver which uses Krylov methods available in Krylov.jl.

There are a few considerations on the design that I think would be helpful to hear your thoughts on:

1. Is there a place for both ConjugateGradientPoissonSolver and KrylovPoissonSolver? Should we keep both methods, or deprecate ConjugateGradientPoissonSolver?
2. There are multiple choices of preconditioners we can choose from, both from Oceananigans and Krylov.jl. For now I have kept the preconditioners in the file src\Solvers\conjugate_gradient_poisson_solver.jl, while putting the new KrylovPoissonSolver in src\Solvers\krylov_poisson_solver.jl which is also compatible with the preconditioners we have. Should we reorganize the files so that we put preconditioners in a separate file? Then perhaps their relations would be clearer.
3. Should we develop a unified interface to allow users to use preconditioners from Krylov.jl as well?
4. I am also not good at writing summaries, so if people have ideas on how to write good Base.summary I am happy to hear your thoughts on what's important!

@glwagner @simone-silvestri @amontoison @tomchor

[amontoison]

I think a dedicated file for the preconditioners is a good idea.

The common API for preconditioners is ldiv!(y, P, x) or mul!(y, P⁻¹, x) (three arguments).
The current API in Oceananigans.jl is precondition! with a variable number of arguments.

It is because the update of the preconditioner and the application of it is fused in Oceananigans.jl.

Related documentation on preconditioners: https://jso.dev/Krylov.jl/dev/preconditioners/

[simone-silvestri]

I would just remove the ConjugateGradientSolver