Add microfluidic device and biology tutorials

[marpaia]

Purpose

This final layer supplies runnable reference compositions for the new wall, flow, device, coupling, and washout APIs. It introduces no new core discretization; its methodological contribution is to make cross-subsystem assembly and scientific assumptions inspectable.

Key methodological contributions

• Consistent device assembly: the trap examples use one device definition to create mechanics walls, a conservative solid mask, fixed media reservoirs, gap-height mobility, and a solved flow field. The pillar example applies the same consistency rule to curved geometry: the smooth mechanics wall encloses the staircase obstacle mask, leaving fluid sites available at every mechanically reachable sampling position.
• Closed phenomenological feedback loop: at a fixed cadence, current cell volumes are rasterized into Brinkman resistance, depth-averaged flow is re-solved, and the live velocity field is replaced. That field then transports signals and kinematically drifts released cells; cells beyond the outlet are removed while lineage is retained. This is a bidirectional model loop, not force-balanced fluid-structure interaction.
• Biological-device compositions: microfluidic_trap.py couples nutrient delivery, consumption, growth, flow, and washout; danino_clock.py places the SimBOL quorum clock and a state-dependent AHL loss field in that device; biopixel_trap.py models one Prindle trapping region; and pillar_channel.py combines cylinder walls, conservative voxelization, adhered founders, released daughters, colony feedback, and washout.
• Evidence-separated Prindle model: the publication supplies the 100 x 85 x 1.65 micrometer trapping region and nominal 500-biopixel context; the supplied CAD supplies an exact 496-outline, 16 x 31 layout and source-specific scale evidence; the adjacent channel, numerical wall thickness, flow, transport, drag, and refresh cadence are labeled model choices. The previous guessed 5-micrometer inset and 100 x 95 cavity are removed. The example validates the CAD layout but does not derive cavity geometry from it.
• Solver-selection guidance: the tutorials distinguish the depth-averaged closure used for shallow-device authoring and frequent feedback from the resolved MAC solver used for wall shear and cross-channel profiles when the mesh resolves the gap.

Scientific scope

• These are qualitative reference models, not reproductions or validations of the Prindle or Danino experiments.
• The biopixel example simulates one trap under one chosen local inlet condition. It does not assert uniform flow across the array, reproduce the array manifold, or include inter-trap coupling.
• The documented CAD scale is an inference corroborated by the publication, not a generic DXF unit convention. The drawing is not asserted to be the publication's exact fabrication revision, and unplaced block geometry is not assigned a fabrication role without a process map.
• The configured channel-scale Péclet number characterizes the model only; no experimental flow or diffusivity measurements are fitted, so no experimental Péclet-number fidelity is claimed.
• Colony blockage uses a center-voxel, Kozeny-Carman-style closure with chosen drag and refresh parameters; cell response remains PR Add experimental kinematic flow coupling #4's kinematic rule.
• Quantitative biological or device claims require measured boundary conditions and material properties, grid and timestep convergence, and parameter sensitivity studies.

Stack

PR 5 of 5.

Base: #5 (marpaia/microfluidics-device-assets)

This is the top of the five-PR stack.

Validation

• Focused mask, device, tutorial, and long-run signal tests: 51 passed
• CPU-only Python suite: 254 passed, 63 skipped
• Metal application suite with the pinned original CellModeller source: 289 passed, 29 skipped; CPU/Metal legacy example matrix passed
• Metal native conformance: 21 of 21 passed
• CUDA 12.8 compile check: all kernels and conformance executables compiled and linked; no NVIDIA runtime was exercised
• Viewer: 13 tests passed; formatting, TypeScript, and production build passed
• Full pre-commit suite, including Ruff, clang-format, and Pyright: passed