test: refresh fuzzy typo probe

[hev]

Closes #13.

Decision

This is stale test truth, not a fuzzy regression. The old local filesystem probe used alpho against alpha. After RFC 0001, FTS runs on unstemmed alyze word_v4 tokens in text_tok; after RFC 0004, fuzzy expansion is length-laddered per analyzed token and 1-5 character tokens intentionally stay exact. alpho is five characters, so the empty fuzzy result is the designed short-token exact path.

The test now uses the RFC 0004 long-token typo probe: kubernets against kubernetes. Exact FTS still returns no rows, while fixed edit distance 1 fuzzy FTS surfaces the document.

Evidence

Failing on main before the test-truth update:

./scripts/cargo test -j 2 -p hevsearch-core --test manager_local_fs local_fs_fuzzy_fts_matches_typo_query -- --exact --nocapture
thread 'local_fs_fuzzy_fts_matches_typo_query' panicked at crates/hevsearch-core/tests/manager_local_fs.rs:599:29:\nindex out of bounds: the len is 0 but the index is 0\ntest local_fs_fuzzy_fts_matches_typo_query ... FAILED\n```\n\nPassing after the update, after rebasing on `origin/main`:\n```\n./scripts/cargo test -j 2 -p hevsearch-core --test manager_local_fs local_fs_fuzzy_fts_matches_typo_query -- --exact --nocapture\ntest local_fs_fuzzy_fts_matches_typo_query ... ok\ntest result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 16 filtered out\n```\n\n## Unblocks\nPR #12 can proceed first, then PR #10 after #12, because both were inheriting this red main test.

[hev]

spec-review (conformance pass): ✅ approve — reviewed against issue #13's decision protocol and RFCs 0001/0004.

• Decision justified: old probe alpho→alpha is a 5-char token; RFC 0004's length ladder intentionally keeps 1–5 char tokens exact, so the old assertion tested the designed exact path, not typo tolerance. New probe kubernets→kubernetes (ED1) is the RFC's own long-token example. This is stale test truth, not a fuzzy regression — wave2's alyze work stands.
• Diff scope: test-only, 2 string swaps + an explanatory comment citing the RFCs. No product code touched.
• Evidence: failing-on-main → passing-after run output present; rebased on origin/main.

Merge order: this first, then #12 (rebase), then #10 (rebase). (Posted by the dispatch loop — the spec-review skill (agentic-gates §5.2) isn't built yet; this is the manual equivalent.)

[hev]

CI attribution (dispatch loop): the red check failed at the cargo fmt step (job 85866916560, step-level conclusion) — i.e. main's pre-existing fmt debt. This PR's own diff is 2 lines in one test file; fmt never reaches clippy/tests on this branch.

#12 and #14 are mutually red through main: #12 (fmt+clippy fixes) was red only on the test this PR fixes; this PR is red only on the fmt debt #12 fixes. Neither can go green alone.

Recommended order (hev's call — a red merge is exactly the admin-judgment case):

1. Merge this PR red-but-explained (smallest diff, failure provably inherited, single-test evidence in body), then
2. Fix manager clippy debt #12 rebases → all three gates green for real → merge green,
3. Add bench evidence CI workflow #10 rebases → green → merge. Main is then fully green with the bench lane live.

[hevbot]

Superseded by #15, which stacks this fuzzy typo test fix with the clippy fix from #12 and the main-branch formatter verification so CI can go green in one PR. Leaving this open for hev to decide on merge/closure.