test: keep two Windows-only flakes off the release gate

[sunerpy]

Summary

Two Windows-only test flakes failed main CI during the v1.6.1 alignment, each in a test the change under test never touched (4231ee5 run 36795226602, 678927c run 36817359513). The release workflow waits for an exact-SHA CI Success on the release commit, so either flake could stall the 0.51.0 release.

• definition_grouping: the temp project was named by pid and nanoseconds only, and the file's parallel tests share the prefix. Windows' clock gave two of them the same nanosecond, so the second init failed with "cannot create an initial index lease because the namespace already exists". A process-wide sequence number keeps the directories apart.
• watcher::tests::root_gitignore_reload_reconciles_and_readmits_sources: waited two seconds for a full reconcile after a .gitignore change, which a loaded Windows runner exceeded (gitignore-removal full sync: Timeout). It now waits ten; the assertions are unchanged.

Test-only; no product or golden impact.

Verification

• Both tests pass locally; the pre-push gate (full suite, 3,903 passed, 0 failed; clippy, fmt, guardrail, docs) passed on the pushed head.

🤖 Generated with Claude Code

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

@@           Coverage Diff           @@
##             main     #283   +/-   ##
=======================================
  Coverage   95.29%   95.29%           
=======================================
  Files         158      158           
  Lines       97623    97623           
=======================================
  Hits        93028    93028           
  Misses       4595     4595           

Files with missing lines	Coverage Δ
crates/codegraph-watch/src/watcher.rs	95.26% <100.00%> (ø)

🚀 New features to boost your workflow:

• ❄️ Test Analytics: Detect flaky tests, report on failures, and find test suite problems.
• 📦 JS Bundle Analysis: Save yourself from yourself by tracking and limiting bundle sizes in JS merges.