MOOSE backend: add runtime parity test against mock RankTwoTensor

[petlenz]

Summary

tests/MooseTargetTest.cpp is purely textual ("does this string appear in the emitted source?"). The compile-check driver only exercises StandaloneCxxTarget. There is no test that:

1. Constructs a MOOSE-like buffer.
2. Runs the generated MOOSE-target compute function through it.
3. Verifies the arithmetic.

Result: the entire cluster of bugs in issues #6 (arg-order), #7 (dim drift), #9 (layout) are undetectable in CI.

Fix

Add tests/moose_mock/ containing a minimal stand-in:

• struct RankTwoTensor { double _coords[3][3]; double const* dataPointer() const { return &_coords[0][0]; } /* + non-const */ };
• struct RankFourTensor { double _data[81]; /* + dataPointer() */ };
• struct MaterialProperty<T> { std::vector<T> _vals; T const& operator[](size_t qp) const { return _vals[qp]; } /* + non-const */ };
• Just enough surface for _qp, MaterialProperty<T>::operator[], dataPointer().

Then:

1. Build a ConstitutiveModel (e.g. linear elasticity).
2. Generate the MOOSE .h + .C via MooseMaterialTarget.
3. Compile and link them against the mock headers.
4. Drive a few QP evaluations and assert the computed stresses match an analytic reference.

This catches arg-order, dim, and layout regressions in one test.

Refs

• REVIEW.md → TEST-A7 (High), CORR-22
• Severity: High
• Depends on: MOOSE call-site argument order diverges from compute function signature #6, MOOSE adaptor ignores tensor dim — silent miscompute when dim != 3 #7, Recipe builders silently accept duplicate add_* calls #9 landing first (otherwise this test would currently fail for the wrong reasons)

[petlenz]

Closed by Phase 5 of the NumSim compiler epic #28 — see #37 (MOOSE state + tangent wiring). The mock-MOOSE fixture this issue asks for is the verification harness for #37's acceptance.