Define conservative cell-occupied transport and executable reference cases

[marpaia]

Closes #13 within its numerical-design and executable-reference scope.

Select an opt-in coarse geometric-porosity model with authoritative extracellular amounts N = epsilon V c, capsule-union occupancy, a declared harmonic face closure, and conservative remapping when cells change accessible storage. Specify intrinsic versus integrated face velocity, geometry/biology/mechanics transaction ordering, accessible sampling/scatter, zero-storage rejection, old-checkpoint conversion, and native implementation requirements. Keep geometric capsule volume distinct from conserved biochemical biomass.

Add a standalone float64 CPU reference and eight executable tests covering the empty/native limit, unequal accessible volumes, partially occupied closed transport, changing/full occupancy, overlapping/wall-adjacent cells, division/removal, boundary/reaction/cell-exchange ledgers, and geometric/spatial/timestep refinement. Update the modeling guide to state the approximation and its limitations.

Validation: all eight reference tests pass, including comparison with actual native CPU backward Euler. Normalized conservation cases have residuals below 1e-12. Targeted strict pyright reports 0 errors/warnings; Ruff format/check and diff checks pass.

This PR does not enable cell exclusion in production or change native kernels/checkpoint schemas. ADR 0025 explicitly scopes subsequent CPU/Metal/CUDA, coupled staging, checkpoint, flow, and refinement work. Independent base: master.

Independent review is complete against the original acceptance criteria and intended review base. The review report records per-criterion evidence, both corrected campaign findings, exact tested heads, landing order and platform limits.

Merged into master in the approved campaign order: #26, #27, #28, #31, #34, #36, #29, #30, #33, #35, #37, #32. The final tree preserves the reviewed implementation and integration regressions. Prerequisites were integrated with merge commits, and the frozen review bases were retained.

Landing validation: 534 Python tests passed (57 environment-dependent skips), 115 viewer tests passed, Ruff lint, strict type checks, viewer build/format checks, and actual Chromium feature/integration checks passed. Pyright retains the existing native-extension source warning. A Windows-only documentation link-test encoding failure was reproduced and fixed with an explicit UTF-8 read; the corrected final PR heads passed Windows CLI/shutdown checks and CUDA compilation. NVIDIA runtime execution remains unverified.