Render matching capsule surfaces and release replaced instance buffers

[marpaia]

Replace overlapping closed-cylinder/full-sphere cells with co-oriented open cylinders and hemispheres whose tangent vertices and normals match. Rod length moves cap centers without stretching the spherical ends, and the selected-cell overlay uses the same geometry, including zero-length spheres. Frame replacement now disposes instance buffers that previously accumulated despite flat geometry counts.

Validation: 35 viewer tests; strict TypeScript/build/format; actual Chromium before/after isolated, moving-colony, near/far and selected-sphere fixtures. Tests cover exact join positions/normals, radius/endpoints for five length/radius pairs and arbitrary orientations, real ray picking, instance colors, stable-ID selection, aligned highlights and zero-length transforms.

Three instanced draws remain for 512 cells. Tracked WebGL buffers remain 18 across repeated frame replacement and return to zero on clear/dispose, versus 390→744 before. Same-fixture median replacement/render timing was 2.2 ms before and 2.1 ms after, with 2.4 ms p95 for both, on Chromium 153 SwiftShader. These are local software-WebGL measurements, not physical GPU performance claims. Higher shared tessellation increases triangles from 258,048 to 294,912.

Committed before/after screenshots and reproducible validation report document the reproduced join artifact, correction, metrics and remaining ordinary silhouette/pixel aliasing. No simulation smoothing or model-state change.

Closes #21.

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.