db: fix splitting implications
How this looks like to the user, is that you end up losing some return
states. In one particular example, Smatch said that xt_mttg_seq_next()
always dereferences *ppos, but actually there is a if statement. We lost
those states.
Originally ->merged states were not leaf states but then when we started
faking states that changed and now we have ->leaf and is_leaf() to mark
leaf states. This code should be using is_leaf() too look for leaf states
instead of checking ->merged.
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>