fix(update): ignore cron/executions.db sidecars, not just the base file

`/*.db-wal` / `/*.db-shm` only match at the checkout root, so on a flat
install `git stash push --include-untracked` still swept
`cron/executions.db-wal` / `-shm` while the scheduler held the WAL-mode
database open (cron.executions._connect opens it via open_db in WAL mode).
The base file stayed put but its WAL vanished under a live writer, so the
next `_connect()` failed with `disk I/O error` (review finding on #111175).

Use `/cron/executions.db*` like the gateway recovery db rule already does,
covering -wal/-shm/-journal and retired-WAL dirs. Every other non-root db
rule in the block already uses the glob. The regression tuple gains the two
sidecars, and the stash test gets the reviewer's repro against the real
`_stash_local_changes_if_needed`: an open WAL connection with one committed
row must still be readable from a fresh connection afterwards.

.gitignore

@@ -188,7 +188,7 @@ docs/superpowers/*
 /gateway/discord_message_recovery.db*
 /sessions/
 /browser-profile/
-/cron/executions.db
+/cron/executions.db*
 /cron/jobs.json
 /cron.pid
 /gateway.lock