Add upwind advection operator builder in matrix_ops

[jc-macdonald]

Summary

matrix_ops.py currently provides a diffusion (Laplacian) operator builder and Crank–Nicolson / implicit-Euler / trapezoidal time-stepping helpers, but has no advection operator builder.

Motivation

McKendrick–von Foerster-style PDE models (e.g., immune-status-structured compartmental models) require a first-order advection (transport) term of the form:

$$\frac{\partial u}{\partial t} + v \frac{\partial u}{\partial x} = \ldots$$

to model continuous waning along a structured axis. The StageOperatorFactory pattern is already set up to produce (L, R) matrix pairs keyed on dt, so an advection builder would slot in naturally.

Proposed scope

1. Add build_advection_matrix(n, dx, velocity, ...) to matrix_ops.py returning a sparse upwind (or flux-limiter) discretisation matrix.
2. Support uniform grids initially; non-uniform grid support can follow from Non-uniform grid Laplacian operator in matrix_ops #45.
3. Return format compatible with existing build_cn_operators / build_implicit_euler_operators helpers.
4. Unit tests covering CFL stability, conservation, and convergence order.

Related

• Non-uniform grid Laplacian operator in matrix_ops #45 — Non-uniform grid Laplacian operator
• Chemostat PDE example: trait-structured population model #47 — Chemostat PDE example (trait-structured population model)
• Parametric parameter forms for axis-aware parameters op_system#50 — Parametric parameter forms for axis-aware parameters