In general, you should not be trying to include such a situation in your proof.
It would make sense to discuss the possibility that something happens, if your proof should be conducted differently when that thing happens. In your example, we might imagine a proof that goes as follows:
Theorem. If $A$ is an upper-triangular $n\times n$ matrix whose columns are linearly independent, then its diagonal entries are nonzero.
Proof. Let $A$ be such a matrix, and suppose for the sake of contradiction that some of $A$'s diagonal entries are zero.
[Some argument that reduces this to the case where $A_{n,n} = 0$.]
Now, we consider two cases. First, suppose that the $n^{\text{th}}$ column of $A$ is a scalar multiple of the $(n-1)^{\text{th}}$ column. In this case, the columns of $A$ are linearly dependent, contrary to our initial choice of $A$. Second, suppose that the $n^{\text{th}}$ column of $A$ is not a scalar multiple of the $(n-1)^{\text{th}}$ column. In this case, [argument we have yet to come up with].
This is a reasonable way to structure a proof. The problem is that in this particular example, it will turn out that [argument we have yet to come up with] does not benefit from knowing that the $n^{\text{th}}$ column of $A$ is not a scalar multiple of the $(n-1)^{\text{th}}$ column. Once we come up with an argument to handle that situation, that argument applies equally well when the $n^{\text{th}}$ column of $A$ is a scalar multiple of the $(n-1)^{\text{th}}$ column, and we could simplify our proof to avoid considering those cases separately.
When looking for a proof, you should not be trying to find special cases that make your task unusually easy, and dealing with them separately. However, if you found a general argument, and it turned out that the general argument does not work when the $n^{\text{th}}$ column of $A$ is a scalar multiple of the $(n-1)^{\text{th}}$ column, then you might treat that situation as a separate case.