Change interface of fem.petsc.create_vector to allow for general list of function spaces

[schnellerhase]

The creation of PETSc vectors relies internally on a list of tuples (IndexMap, bs). Given a list of such and a vector kind we can create the corresponding coefficient vector. Previously this functionality was not exposed to the user directly and only supported indirectly by passing a Form from which the spaces would then be extracted.

• Introduces la.petsc.create_vector which now takes a list of (IndexMap, bs) tuples and returns a PETSc.Vec object. The interface of fem.petsc.create_vector is extended to allow for function spaces as input as well. Vectors can therefore now be create by a list of function spaces: b = dolfinx.fem.create_vector([V, W]) and no longer require a form to be present for such construction. And alternatively from a general sequence of index maps and blocks sizes, without the presence of any function spaces with la.petsc.create_vector.
• Introduces _assign_block_data which factors out the attachment of block information to PETSc vectors.
• To keep the interface tight it is no longer possible to call create_vector on a form. Enforces explicitness, that the space is the critical argument, not the form.

[jorgensd]

Shouldn't then this also be done for matrices to stay consistent, i.e. have an extract_function_spaces(bilinear_form), and create_matrix(nested_list_of_spaces)?

[schnellerhase]

Shouldn't then this also be done for matrices to stay consistent, i.e. have an extract_function_spaces(bilinear_form), and create_matrix(nested_list_of_spaces)?

Sounds good - lets do that in another PR.

[schnellerhase]

92d1a1b adapts to very fresh ruff release https://pypi.org/project/ruff/0.13.0/

[michalhabera]

Shouldn't then this also be done for matrices to stay consistent, i.e. have an extract_function_spaces(bilinear_form), and create_matrix(nested_list_of_spaces)?

Sounds good - lets do that in another PR.

Is that possible? We inspect type of integrals present in the form to prepare the sparsity pattern of the matrix, see fem::build_sparsity_pattern.

[schnellerhase]

True matrix is integral, and thus form, dependant. If one was to switch basing the construction of a matrix on spaces, creation and sparsity pattern setting would need to be separated. That could be desirable for customisable sparsity patterns, ref #3779 (comment).