The confusion may be just the same as in your question about completeness:
The i.o.w. part of the definition has as its antecedent all the contradictory sentences that can possibly be built, not the ones that are already provable:
Definition #2:
[...] if of any two grammatical sentences that contradict each other, at least one is not provable
or equivalently.
Definition #2:
[...] if of any two sentences in the language that contradict each other, at least one is not in the theory
Reformulating the first variant of the definition,
Definition #1:
[...] if no two asserted statements of this theory contradict each other
$\Leftrightarrow$ if for any two sentences such that both are provable in the theory, they do not contradict each other
it becomes obvious that the one is just the contaposition of the other:
$P = $ both sentences are provable in the theory,
$\neg P = $ at least one sentence is not provable in the theory
$Q = $ the sentences do not contradict each other,
$\neg Q = $ the sentences contradict each other
Definition #1 = $P \to Q$
Definition #2 = $\neg Q \to \neg P$
Since contrapositive statements are logically equivalent, the two formulations mean the same thing.
But you are probably overthinking matters; the inutition behind consistency is quite simple: It just means that the theory doesn't assert anything contradictory, i.e. it never proves $\phi$ and $\neg \phi$ at the same time.