feat: lazy retention sweep on /web/cases

New module retention::sweep_user_retention is invoked at the top of
handle_my_cases, using the current user's retention policy. Per case:
step 1 auto-closes when the newest .m4a mtime is older than
auto_close_days; step 2 auto-purges when closed_at is older than
auto_delete_days. Order matters - a case auto-closed in the same sweep
has closed_at ~= now and naturally survives the purge check,
preserving the full grace period after a vacation.

Race guard: purge re-reads the close marker immediately before
remove_dir_all so a concurrent upload that reopens the case cancels
the purge. Every automatic action emits an info! line with reason and
threshold; failures warn but never abort the sweep.

Seven integration tests cover both threshold paths, the disable-via-0
escape hatch, the newest-recording rule, the vacation scenario, and
the open-case skip. Tests age the comparison objects (mtime via
filetime, closed_at as handwritten RFC3339) instead of faking "now" -
no clock abstraction needed.

server/src/routes/user_web.rs

@@ -312,6 +312,15 @@ pub async fn handle_my_cases(
     auto_trigger::try_enqueue_all_for_user(&user_root, &pipeline.analyze_tx, &config, &events_tx)
         .await;
 
+    // Lazy retention sweep: at most one disk scan per /web/cases visit,
+    // reusing the user's retention policy from users.toml. Runs before
+    // the listing scan so any auto-close/auto-purge actions are
+    // reflected in the same response.
+    if let Some(u) = config.users.iter().find(|u| u.slug == user.slug) {
+        crate::retention::sweep_user_retention(&user_root, &user.slug, &u.retention, &events_tx)
+            .await;
+    }
+
     let busy = pipeline.analyze_busy.0.load(Ordering::Acquire);
     let cases = scan_user_cases(&config, &user.slug, busy).await;
     let total = cases.len();