The timeout value was added to the command twice, which is incorrect. The correct position is after the keys; inserting it before the keys makes the timeout value a key itself (one that likely does not exist).

That doesn't need to be especially problematic with a standalone client, because Redis will just ignore keys that do not exist, but it becomes especially problematic with a cluster client, because the command may suddenly target multiple nodes. This ends up with an error immediately, on the client side.

Fixes #48585