fix(tests): make unwritable-path tests actually unwritable on Windows (V2-1043)

[Nic-dorman]

Linear issue

V2-1043

Risk tier

• T0 — docs / tooling / CI / pure UX-output. Repo CI only.
• T1 — client-only, no network-facing behavior change. CI + prod compat smoke.
• T2 — node/client logic with behavioral surface, no protocol/format/economics change. Dev testnet + ADR.
• T3 — protocol / storage format / payments / routing. T2 evidence + adversarial testing.

Test-only change; no shipped code touched.

Compatibility

• Wire: none
• Storage: none
• API: none

Semver impact

• breaking
• feature
• fix

Test evidence

Both tests assumed a path under the Unix root is uncreatable; on Windows it resolves to a creatable C:\ directory, so save_snapshot_to_unwritable_dir_does_not_panic failed on every Windows dev box and save_with_timeout_returns_promptly_on_fast_failure silently timed a successful write instead of a fast failure — and both littered directories at the drive root. The fix routes the path through a tempfile::NamedTempFile so a parent component is a regular file and create_dir_all fails on every platform.

• Windows 11 (previously failing environment): both tests pass; full cargo test -p ant-core --lib now fully green, 452/452; confirmed no drive-root litter is recreated.
• cargo clippy --all-targets --all-features -- -D warnings clean, cargo fmt --check clean.
• Linux/macOS behavior unchanged (path is uncreatable there under either construction); repo CI covers it.

New dependency

none (tempfile is already a dev-dependency used by neighboring tests)

ADR

n/a

Mitigation / rollback

Test-only — revert the single commit.

🤖 Generated with Claude Code

[dirvine]

Review verdict: APPROVE at c437471b025ff876cbb9035e06fbb57adc4f9af2.

Routing the failure through an existing regular file is deterministic across Windows/Unix and actually exercises the intended create_dir_all error path without relying on Unix permission semantics. The focused local test passed: save_snapshot_to_unwritable_dir_does_not_panic — 1 passed; the PR's Windows CI is also green. I found no diff-specific blocker.

The remaining red checks are the repository-wide h2 security audit and the pre-existing Ubuntu Merkle E2E timeout/cancellation.

[jacderida]

Approved. Verified the diff is confined to test code inside #[cfg(test)] — no shipped code is touched; routing the path through a NamedTempFile makes create_dir_all fail on every platform rather than relying on Unix root-write semantics.

The two red checks on this branch are the known repo-wide baselines, both already fixed on main: Security Audit was RUSTSEC-2026-0258 (h2 0.4.14), cleared by #177, and Merkle E2E (ubuntu) was killed at the 60m cap, raised to 90m by #178. Neither is attributable to this diff.