To avoid a potential race with the key zeroing modifying
the next hash key retrieved by the loop in `Table.zero()`,
retrieve all the keys in user space first before starting
the zeroing loop. See discussion on #780.

Tested with `funccount 'SyS_*' -i 1` while running a heavy
read/write test application (`dd`) in the background for
several minutes with no visible issues.