SNES solver interface for standard, block and nest systems

[jorgensd]

A "functional" approach to the snes problem interface.
This reduced the code to:

• One factory function: create_snes_solver, that returns the the user a SNESObject and a vector to solve into x
• A SNESSolver, that handles destruction of PETSc objects. It also store convenient wrappers to transfer data from your dolfinx.fem.Function|list[dolfinx.fem.Function (u) to the PETSc vector used for solving (x)

The tests now show examples of:

1. How to use SNESSolver, the highest level interface
2. How to use create_snes_solver, the mid-level interface
3. How to use the functions supplied in dolfinx.nls.petsc and dolfinx.fem.petsc to interact with your own snes solver instance.

[jhale]

My impression is that we should first tackle the comment made by @garth-wells https://github.com/FEniCS/dolfinx/pull/3648/files#r1973997836 in a separate PR.

This can then be refactored to use #3659 which should remove many of the if statements regarding assigning functions to vectors and vice-versa. We can then think about if there are any smart ways to deal with Mat and Vec types in the SNES solver setup.

[jhale]

I propose I code some further adjustments:

• API change: SNESSolver immediately becomes NonlinearProblem, i.e. a high level class that wraps SNES, as LinearProblem is a high level class that wraps KSP.
• API change: Current NonlinearProblem becomes NewtonSolverNonlinearProblem and is marked as deprecated.
• API change: Python NewtonSolver is marked as deprecated.
• compute_jacobian has name changed and documentation string added.
• Consider further factoring into private functions in assemble_residual.