test: cover unavailable cwd after successful lookup

[hyfdev]

Summary

• add a Unix child-process test that successfully initializes ambient cwd state, removes the real cwd, and distinguishes default lookup errors from the cached_current_dir snapshot
• compare dependent Path, str, and String failures with the direct current_dir error, and prove explicit absolutization remains independent of ambient cwd
• require the lifecycle test in Linux and macOS CI under both cwd configurations and record the covered contract

Why

Existing tests covered an unavailable cwd before the first successful lookup, retry after that failure, and cwd changes between valid directories. They did not detect a cached provider that re-read or revalidated the real cwd after a successful cache initialization.

A controlled mutation that calls current_dir() on a cache hit survives the previous cwd tests but fails the new lifecycle test after the directory is removed.

Impact

This is a test-only contract change. Production code, dependencies, benchmarks, and allocation baselines are unchanged.

Validation

• cargo fmt --all --check
• SUGAR_PATH_EXPECT_CACHED_CURRENT_DIR=0 cargo test --locked -p sugar_path
• SUGAR_PATH_EXPECT_CACHED_CURRENT_DIR=1 cargo test --locked --workspace --all-features
• targeted default and cached_current_dir runs for all affected integration tests
• cargo clippy --locked --workspace --all-targets --all-features -- -D warnings
• RUSTDOCFLAGS="-D warnings" cargo doc --locked --workspace --all-features --no-deps
• default/cached test registration and ignored-test checks
• independent adversarial review with no findings