test(applied): pin block-transposition adjacency in every shared layer (#373)

[tsondru]

Closes #373.

Adds review_soundness_fixes::block_transposition_needs_adjacency_in_every_shared_layer to catgraph-applied/tests/smc_nf_regression.rs: a fixture where a braid-carrying component separates two free components' runs in 2 of their 3 shared layers, asserting the exact nf() output. Production diff: 0 lines (git diff --stat 27be055..HEAD: tests/smc_nf_regression.rs +92, CHANGELOG.md +7, docs/SMC-NF-RECONCILIATION.md ±1).

Premise re-scoped at the STOP (owner call, recorded on #373): at 27be055 the b1 == a2 deletion reddens 6 tests in debug and 4 in release via corpus statistics; this test is the narrow, targeted oracle.

Falsification (closing review, Sonnet rust-v2:falsifying-reviewer, /tmp/audit-reports/G5-T6/close-review.md)

#	perturbation	profile	named test	observed
1	delete b1 == a2 in adjacent_runs (smc_nf.rs:2148)	debug	block_transposition_needs_adjacency_in_every_shared_layer	FAIL at the pre-existing debug_assert! smc_nf.rs:2041 (30/31 pass)
1	same	release	same	FAIL at the test's own assert_eq! (smc_nf_regression.rs:772): [Identity(1),Braid(1,1)],[G,G,Eta,F],[Eps,Eps,F,G],[Eps,Identity(1)] vs expected [Identity(1),Braid(1,1)],[F,G,G,Eta],[G,Eps,Eps,F],[Identity(1),Eps] (30/31 pass)
2	blocks_are_adjacent decides on the first shared layer only	debug + release	same	PASS — null: this fixture's first shared layer already fails adjacency, so the "every layer" requirement is not distinguished from "first layer" by this fixture

Round 1 (lane, Opus) ran on 2f07e0c and its findings were applied; the closing review found one minor (a backtick span wrapped across a line in the CHANGELOG bullet), fixed in the final docs-only commit, which is unreviewed.

Gates (pristine copy, closing review)

cargo nextest run -p catgraph-applied: 670 passed / 0 failed / 12 skipped. cargo clippy -p catgraph-applied --all-targets -- -D warnings: clean. cargo fmt --all --check: clean.

🤖 Generated with Claude Code