Phase 2a: emit strain-coupled implicit-residual materials (Mode B)

[petlenz]

What

Teaches NumSimMaterialTarget to emit strain-coupled implicit-residual materials — a recipe whose scalar state is defined by an implicit R(x, ε)=0 (via add_scalar_residual_equation, landed in #99) is lowered to a numsim-materials material that holds a material_ref<backward_euler> and drives the Newton loop itself in one compute() (Mode B, the robust pattern small_strain_plasticity uses — no Local-edge ordering fragility). The consistent tangent dσ/dε is deferred to PR 2b and rejected loudly here.

Commits

1. 90f98d9 — Finding A guard. Before this PR, a residual recipe was mishandled on main: NumSimMaterialTarget threw a misleading rate/rk_integrator error, and the self-contained path (emit_compute_function, standalone/MOOSE) silently emitted a function dropping the state. Now both reject loudly with an accurate message (the self-contained guard is permanent; the target one is replaced by emission in commit 2).
2. f6edc2a — Mode-B emitter. emit_residual_material(): material_ref<backward_euler> + solve(eval) returning {residual, jacobian} (∂R/∂x via cas::diff), state-apply, stress. emit() routes residual recipes here. Unit tests + a gcc-gated cross-repo e2e against the real backward_euler (z=c·tr(ε), σ=z·ε).
3. df9fb8e — review fixes (3-lens critical review folded in). See below.

Verification

• 294/294 tests pass (g++-14), including the gcc-gated cross-repo e2e driving the real solver.
• The emitted material was instantiated & run end-to-end; multi-output headers template-instantiate against real numsim-materials.
• The nonlinear-residual e2e validates the emitted jacobian (sabotage-checked: forcing jacobian=1.0 makes it fail).

Review fixes (commit df9fb8e)

Four HIGH emitted-code-correctness bugs the single-output test recipe never exercised: multi-output CSE temp collision (t0 redeclared), <state>_old → unbound identifier, solver/eval/residual/jacobian name collisions, and an e2e that couldn't catch a wrong jacobian (linear ∂R/∂z≡1). All fixed + pinned by tests.

Deferred (tracked): backward_euler::solve() clamps std::max(x,0) (plasticity convention — documented in the emitted header; upstream issue for an unclamped mode to follow); ~445-line duplication vs the rate path (extract detail:: helpers after PR 2b).

🤖 Generated with Claude Code