applied: MatR::zero_matrix contents are unpinned — filling one() instead of zero() reddens nothing

[tsondru]

Provenance: G4-T4 (#301) review, 2026-08-27. Not introduced by that diff.

Perturbing MatR::zero_matrix (src/mat.rs:77) to fill R::one() instead of R::zero() reddens nothing workspace-wide (146 binaries, 0 failed). Its call sites read shape or a property the fill does not change: the round-trip oracle is S(mat_to_sfg(M)) == M for whatever M the constructor produced; try_inverse_of_singular_is_none (tests/mat_f64.rs:178) holds for an all-ones 2×2 too; mat_kron.rs:273 and trace.rs:139 use it for shape.

Fix: pin the entries directly — zero_matrix(2, 3).entries() all R::zero(), on at least one rig whose zero is not 0.0 (Tropical::zero() is +∞).

MatKron::zero_matrix (src/mat_kron.rs:82) delegates and inherits the gap.

[tsondru]

Folded into #392 (coverage backlog, 2026-09-01 triage: test-only row, prod diff 0 expected). Reopen if a production defect surfaces here.