fix: make main CI green (stacked fmt + clippy + fuzzy-test fixes)

[hevbot]

What was broken

Main CI was red in three independent slices, while sibling PRs each fixed only one lane:

• test: refresh fuzzy typo probe #14 (fix/fuzzy-typo-test-truth) fixed the fuzzy typo test truth but failed at the fmt stage.
• Fix manager clippy debt #12 (fix/clippy-debt-manager) fixed the manager.rs clippy -D warnings debt tracked by main fails its own CI: clippy -D warnings violations in manager.rs (blocks PR #10 / issue #7) #11 but failed at the fuzzy test stage.
• Add bench evidence CI workflow #10 (feat/bench-evidence-ci) still fails at the clippy stage until the main-line clippy debt is cleared.

What this stacks

• Cherry-picks Fix manager clippy debt #12, preserving Adam Hevenor as author: Fix manager clippy debt.
• Cherry-picks test: refresh fuzzy typo probe #14, preserving Adam Hevenor as author: test: refresh fuzzy typo probe.
• Runs the formatter against the stacked main branch. ./scripts/cargo fmt produced no remaining diff after the two cherry-picks, and ./scripts/cargo fmt --check passes. No Add bench evidence CI workflow #10 fmt-only commit was cherry-picked.

Closes #11.
Closes #13.

After this lands, #10 should rebase onto a green main CI baseline.

Local verification

./scripts/cargo fmt --check
# passed

cargo clippy --workspace --all-targets -- -D warnings
# Finished `dev` profile [unoptimized + debuginfo] target(s) in 1m 16s

./scripts/cargo test -j 2 -p hevsearch-core --test manager_local_fs local_fs_fuzzy_fts_matches_typo_query -- --exact --nocapture
# test local_fs_fuzzy_fts_matches_typo_query ... ok
# test result: ok. 1 passed; 0 failed; 0 ignored; 0 measured; 16 filtered out; finished in 0.19s

[hevbot]

Follow-up after the first #15 CI run:

• Remote cargo fmt passed.
• Remote cargo clippy --workspace --all-targets -- -D warnings passed.
• Remote cargo test --workspace --no-fail-fast passed.
• Remote ignored MinIO integration tests then failed in service_semantic_cache / service_result_projection because those semantic-cache fixtures created default single-vector namespaces, now L2, then opted into semantic cache.

Pushed 9deb022 test: seed semantic-cache fixtures with cosine metric, which makes those semantic-cache test preconditions explicit by seeding the fixtures with DistanceMetric::Cosine. Local ./scripts/cargo fmt --check and cargo clippy --workspace --all-targets -- -D warnings pass after the patch. A scoped local MinIO rerun could not validate assertions because the existing local service on 127.0.0.1:9000 rejected the repo/CI default minioadmin credentials with InvalidAccessKeyId; CI starts a fresh MinIO with those credentials.