[FEA] Enforce input types across the routing Python API (type hints + centralized validation)

Is your feature request related to a problem? Please describe.

The routing Python API (DataModel setters, Solve, SolverSettings, Assignment) does not enforce input types consistently:

• Historically, inputs had to be cuDF but were only checked incidentally — e.g. a numpy/pandas input failed with a cryptic AttributeError: 'numpy.ndarray' object has no attribute '__cuda_array_interface__' rather than a clear type error.
• PR #1546 added numpy/pandas support via shared helpers (type_cast / prepare_matrix) that normalize numpy/pandas/cuDF and reject Python lists — but enforcement is per-setter and easy to bypass. That work uncovered set_order_prizes silently bypassing the shared type_cast (it called .astype() directly), which only surfaced through a comprehensive cross-backend test, not any type check.
• There are no PEP 484 type annotations on the public API, so IDEs/mypy cannot catch wrong types, and error messages are inconsistent.

Describe the solution you'd like

Consistent, centralized type enforcement for the routing Python API:

1. Route every array/scalar parameter through one shared validate-and-normalize helper — no setter may access the device pointer without going through it.
2. Add PEP 484 type hints to the public classes/methods (DataModel, SolverSettings, Solve, Assignment) declaring accepted types (numpy.ndarray, pandas.Series/DataFrame, cudf.Series/DataFrame), checkable by mypy/IDEs.
3. Fail fast with a consistent TypeError/ValueError on unsupported types (e.g. Python lists/tuples), wrong dtype, or wrong shape.
4. Add a test/lint guard that fails if a setter dereferences __cuda_array_interface__ / builds a device pointer without going through the shared helper — preventing future set_order_prizes-style bypasses.

Describe alternatives you've considered

• Docstring-only documentation of accepted types (current state) — not enforced, no tooling support.
• Per-setter ad-hoc isinstance checks — inconsistent and prone to drift (this is what caused the set_order_prizes gap).
• Runtime validation decorator vs. typing.get_type_hints introspection for the enforcement mechanism — to be decided during design.

Additional context

• Related PR: #1546 (numpy/pandas input support in routing DataModel).
• Shared helpers today: type_cast / prepare_matrix in python/cuopt/cuopt/routing/vehicle_routing_wrapper.pyx; validators in python/cuopt/cuopt/routing/validation.py.
• Consider extending the same policy to the LP/MILP/QP (numerical optimization) Python API for API consistency.