To-think: a less restrictive type for the type of the operator in LinearSolver abstract interface?

[amartinhuertas]

To-think (consequences among others)

The current LinearSolver abstract interface in Gridap forces the type of the operator to be AbstractMatrix. As we want to accommodate a larger zoo of linear solvers within such interface, such as, e.g., Krylov subspace solvers, this might be too restrictive, essentially because a linear operator in a Krylov subspace solver essentially requires a very reduced set of operations in the interface of AbstractMatrix.