test(picker/summary): sleep between writes to avoid mtime tick collision (#2433)

## Problem

`test_cache_roundtrip_and_prune_on_write` flaked on PR #2429's
`code-coverage` job
([logs](https://github.com/max-sixty/worktrunk/actions/runs/24975304781/job/73126010174))
— the assertion that `deadbeef` was pruned after writing `cafebabe`
failed, then a rerun passed.

Root cause: `sweep_lru` (`src/cache.rs:126`) sorts entries by file mtime
to pick the eviction victim. The test wrote `deadbeef` and `cafebabe`
back-to-back with no delay; on filesystems with coarse mtime resolution
both files can share a tick, making eviction order nondeterministic and
sometimes pruning the newer entry instead.

## Solution

Add a 10 ms sleep between the two writes — the same gap
`test_sweep_lru_trims_oldest_entries` (`src/cache.rs:319-322`) already
uses for the same reason.

## Testing

Targeted test passes locally; the broader
`commands::picker::summary::tests` and `cache::tests` modules still
pass.

---

Closes #2432 — automated triage

Co-authored-by: worktrunk-bot <254187624+worktrunk-bot@users.noreply.github.com>
Co-authored-by: Claude <noreply@anthropic.com>

Author: worktrunk-bot

src/commands/picker/summary.rs

@@ -142,6 +142,12 @@ mod tests {
         // Different hash for the same branch is a miss (content-addressed).
         assert!(CachedSummary::read(&repo, branch, "cafebabe").is_none());
 
+        // sweep_lru picks the eviction victim by file mtime; on filesystems
+        // with coarse mtime resolution, back-to-back writes can share a tick
+        // and the order becomes nondeterministic. Match the gap used in
+        // test_sweep_lru_trims_oldest_entries.
+        std::thread::sleep(std::time::Duration::from_millis(10));
+
         // Writing a second hash for the same branch prunes the old one.
         let second = CachedSummary {
             summary: "Refactor tests".to_string(),