engine/tests/{bc_validation,shell_validation}.cpp exist and pass but are never run in CI

[mkofler96]

Problem

engine/tests/bc_validation.cpp (113 lines, 17 cases) and engine/tests/shell_validation.cpp (296 lines, 3 top-level test functions) are real native regression tests with their own build scripts (scripts/test-bc-validation.sh, scripts/test-shell.sh). Both compile with a plain host clang++/g++ — no MFEM/OCCT/Netgen/Emscripten dependency — specifically so they can run as a fast unit-test loop independent of the WASM build.

Neither script is referenced anywhere in .github/workflows/ci.yml (grep -n "test-bc-validation\|test-shell\|bc_validation\|shell_validation" .github/workflows/ci.yml returns nothing). The only reference to either is a mention in examples/shell-coupling/README.md:39.

I ran both locally to confirm they're not just stale/broken placeholders:

• bash scripts/test-bc-validation.sh — 17/17 checks pass, sub-second. Covers exactly the "reject out-of-range vertex/DOF indices loudly" contract from solve_mfem.cpp silently drops out-of-range fixed/prescribed DOFs and point loads instead of erroring #362 and the shell DOF-component validation from solve_shell.cpp silently drops out-of-range vertex/DOF indices in fixed DOFs, fixed vertices, and point loads (shell-solve equivalent of #362/#367) #379.
• bash scripts/test-shell.sh — all checks pass in ~2.9s. Covers DKT+CST shell element math and the "constraint on dependent RBE3 node throws" contract from Fixed-DOF constraints on shell RBE3 coupling reference nodes are silently dropped #377.

Impact

These two harnesses are the only tests that directly lock in the correctness fixes from #362, #377, and #379 (all "silent fallback → loud error" fixes that were explicitly called out as important in this project's ongoing hardening effort). Because they're never invoked by CI, a future change to engine/cpp/bc_validation.h or engine/cpp/shell_core.cpp that regresses that behavior would merge cleanly — none of the four CI jobs (rust, clang-tidy, wasm, frontend) would catch it. The Playwright suite exercises the WASM build end-to-end but doesn't target this validation logic directly, and cargo test --workspace covers the (currently test-less, see #365) Rust crates, not engine/cpp.

This is a coverage gap distinct from #365 (native Rust bridge crates have zero tests) — these C++ tests exist, pass, and are cheap; they're just not wired up.

Suggested fix

Add a step to the clang-tidy or a new lightweight job in .github/workflows/ci.yml that runs both scripts, e.g.:

- name: Native validation tests
  run: |
    bash scripts/test-bc-validation.sh
    bash scripts/test-shell.sh

Both run in well under 5 seconds combined and need no WASM toolchain, so this is close to free CI time — much cheaper than adding another Playwright spec for the same coverage.

[mkofler96]

we should think about adding something like https://github.com/google/googletest or catch2