test(applied): value-pin mat_f64::determinant at n = 0..4; CI f64-rig lane (#296)

[tsondru]

Closes #296. Taskmap G3-T4 — closes G3.

Measured at 625ecb1: determinant returning constant Some(1.0) passed 45/45 applied test binaries (--features f64-rig --no-fail-fast); tests/mat_f64.rs 8/8 green. No CI lane compiles tests/mat_f64.rs (rg 'f64-rig' .github/workflows/ empty).

Shipped (production diff 0 lines; src/mat_f64.rs change is rustdoc only):

• tests/mat_f64.rs: check_determinants helper (collects every mismatch, reports measured values) + #[should_panic] self-test; value pins at n = 0 (→1), 1 (7, −7), 2 (6, 0, −1), 3 (58, −1), 4 (4, −1, 0). Expected values hand-derived before running; the reviewer re-derived all independently (block-diag measured 4.000000000000011, LU error 1.1e-14).
• src/mat_f64.rs, examples/mat_operations.rs: every "LU" attribution deleted — nalgebra 0.35 determinant is closed-form for n ≤ 3 and try_inverse for n ≤ 4 (src/linalg/{determinant,inverse}.rs); rg LU over both files is empty.
• ci.yml: cargo test -p catgraph-applied --features f64-rig + clippy --all-targets -D warnings. Unscoped, matching the serde lane at :122 (226 lib + 448 integration + 7 doc tests).
• CHANGELOG bullet.

Falsification (--test mat_f64, 13 tests):

perturbation	red	message values
Some(1.0)	1x1_and_3x3, 2x2, 4x4	7→1, 58→1, 6→1, 0→1, 4→1, −1→1, 0→1
Some(det.abs())	1x1_and_3x3, 2x2, 4x4	−7→7, −1→1 (I₃), −1→1 (2×2), −1→1 (I₄)
product of diagonal	1x1_and_3x3, 2x2, 4x4	58→18, 0→4, 4→160, −1→0
Some(-det)	all five value tests + identity	1→−1, 7→−7, 58→−58, 6→−6, 4→−4, −1→1
guard != → ==	all five + non_square	None on every square fixture
helper assert! dropped	check_determinants_reports_a_mismatch	"did not panic"
Some(f64::NAN)	all five value tests + identity + self-test	every fixture "measured NaN" (before 683e1cc the four check_determinants tests were green; only the identity pin and the self-test caught it)

Nulls: transpose-then-det (det Aᵀ = det A); tolerance >= → > (no fixture on the boundary); row1 = 2·row0 under abs() and diagonal-product (0 → 0); determinant_0x0_is_1 under constant-1 and diagonal-product (empty product = 1). Pre-existing determinant_of_identity_is_1 is green under constant, abs, and diagonal-product — the hole this closes.

Reviews: r1 executing adversarial (0 blocking / 3 important / 6 minor) → r2 executing delta on the applied edits (2 important / 1 minor: NaN-unsafe predicate, a surviving LU gloss on the inverse) → r3 executing on the fix commit 683e1cc (0/0/3 minor). All findings applied except one note-only item: the should_panic substring measured 7 would also accept 7.5; the fixture is an exact 1×1, so no reachable value does.

Out of scope, recorded: examples/mat_operations.rs is self-checking (asserts det 58) but never run in CI — its later sections exceed 600 s in debug; try_inverse already has a value oracle (inverse_matmul_original_is_identity), not widened; #269 sequenced behind this PR.

🤖 Generated with Claude Code