When you initially set the eigenvalue at the top rung to $\hbar l$, you don't need to assume that $l$ is an integer, you can think of it as any multiplicative constant. Clearly there is no loss of generality there. The beautiful aspect of the ladder operator approach is that you can use it to prove that $l$ must be a non-negative integer or half-integer.
This argument is presented clearly in Griffiths, at least in the second edition (perhaps you are using the first edition?). Using the ladder operators $L_+$ and $L_-$, and the conditions that there must be a top rung and a bottom rung for the ladder of eigevnalues, you automatically find that
the eigenvalues of $L_z$ are $m \hbar$, where $m$ ... goes from $-l$ to $+l$ in $N$ integer steps. In particular, it follows that $l = -l + N$, and hence $l = N/2$, so $l$ must be an integer or a half-integer.
So, the nature of $l$ is discovered as a conclusion - there is no initial assumption.