Complete Phase 2 with a convergent staggered-grid solver

[Kandil2001]

What changed

• replaces the production collocated pressure-correction path with a staggered Marker-and-Cell solver
• uses compatible face-velocity divergence and pressure-gradient corrections
• adds strict convergence based on velocity updates, divergence Linf/L2, mass balance, pressure convergence, and consecutive passes
• adds continuation across Reynolds numbers for study modes
• retains RBGS and RBSOR pressure solvers and upwind/central convection options
• adds canonical, medium, grid, and Re=1000 run helpers
• builds the production executable through CMake
• adds the production canonical case to CTest
• adds a standalone CI regression that checks convergence, pressure failures, divergence, velocity residual, and Ghia thresholds
• updates the README and methodology for the verified solver

Verification performed locally

• canonical N=32, Re=100, upwind, RBSOR: converged
• all six N=32 cases at Re=100/400/1000 with upwind and central RBSOR: converged and benchmark thresholds passed
• N=16/32/64, Re=100, central RBSOR grid sequence: converged and benchmark thresholds passed
• canonical runtime was approximately 1.2 seconds on the development environment

Important scope note

The complete 36-case mode is available but includes slower RBGS and N=128 cases. The CI acceptance set intentionally uses the canonical case plus the independent operator, Poisson, and convergence tests.

Main branch

The previously merged verification foundation remains intact. This PR contains the production integration and final Phase 2 documentation.