Revisiting type definitions

[sshin23]

As mentioned in #58 (comment), several major types in MadNLP include non-concrete types. Because of this, several auxiliary functions are introduced to avoid unnecessary memory allocations. Such introduction of auxiliary functions can be avoided by always using concrete types in the type definition. The following is suggested:

• Thoroughly revisit the type definitions (NonlinearProgram, Solver, SparseMatrixCOO, etc) and eliminate the abstract types in the type definition.
• Eliminate the auxiliary functions whenever there's no performance regression.
• Change Solver -> InteriorPointSolver

[frapac]

Closing this issue, as the types in MadNLP have been revamped in the recent refactoring #272