persist_parsed_path_table asserts a wall-clock floor and fails on a fast machine #1028

unchanged_second_run_reuses_the_cache_instead_of_reparsing in packages/rust/tests/persist_parsed_path_table.rs:156 fails on clean main:

the corpus must be big enough for the ratio assertion to mean something;
the cold run took 2.282013488s — raise CORPUS

The test asserts cold >= Duration::from_secs(3) to guarantee the later cold/warm ratio assertion has headroom. That floor is calibrated to the CI runner. On a faster machine the cold run finishes under it and the test fails — not because the cache behavior regressed, but because the machine did the work too quickly.

A wall-clock floor makes the suite's outcome a function of host speed, and it fails in the direction of "your machine is fast", which reads as a real failure to anyone running cargo test --workspace --features cli locally. Better to assert the invariant directly (the warm run parses zero files, which fx.parses() already gives) than to gate it behind a duration.

Found while running the Rust suite for #1002; reproduced on a clean main worktree.