test(ck-engine): guard empty-corpus lexical rebuild

[ak2k]

Test-only, no production change.

A --lex search over a directory with no indexable files builds an empty tantivy index. When real files appear later the corpus fingerprint changes, so the next search must rebuild rather than trust the empty stub — otherwise that directory reports no matches indefinitely. The rebuild logic is already in lexical_search; this pins the empty-corpus corner of it, which is the case I actually hit.

Verified non-vacuous: dropping the fingerprint comparison from the freshness check (let is_fresh = tantivy_index_path.exists();) fails the test with empty tantivy stub was trusted; expected a rebuild to find the new file.

Uses a unique TempDir, which keys its own index, so it needs no serial_test or env manipulation to stay isolated. cargo test -p ck-engine (41), clippy, fmt clean.

[ak2k]

CI is red here, but the two failures are in ck-cli/tests/integration_tests.rs — test_add_single_file_to_index ("Failed to create initial index") and test_add_file_with_relative_path — and this PR only adds a test to ck-engine/src/lib.rs. No production code, nothing in ck-cli.

What I checked locally, on Linux and macOS, using CI's exact invocation (cargo test --manifest-path ck-cli/Cargo.toml --all-features):

• pristine main (f4caaf5): both pass
• this branch: both pass

So I can't reproduce it, and I don't think it's caused by this change. Both tests are #[serial] and shell out to the ck binary to build an index, which is the usual shape for load-dependent CI flakiness — but that's a guess, not something I've confirmed.

I don't have rerun rights on this repo. If you re-run the job and it's still red, tell me and I'll dig in properly rather than assume it's environmental.