gh-47: update wal/oom.test

Tuples are got a big bigger with shared-arena patch (4 bytes
per tuple overhead).

So we exhaust the current arena in out-of-memory test a bit
faster than before.