[guided]The only possible source of divergence in the Petersson integral is the cusp $y\to\infty$. Away from the cusp, the fundamental domain is compact, so ordinary continuity will control the integrand. Near the cusp, we use exactly the condition that $f$ and $g$ are cusp forms.
Define the holomorphic map
\begin{align*}
q:\mathbb{H}&\to\{w\in\mathbb{C}:0<|w|<1\}\\
z&\mapsto e^{2\pi iz}.
\end{align*}
The cusp $y\to\infty$ corresponds to $q(z)\to 0$, because for $z=x+iy$ we have
\begin{align*}
|q(z)|=|e^{2\pi ix}e^{-2\pi y}|=e^{-2\pi y}.
\end{align*}
Because $f$ and $g$ are [cusp forms](/page/Cusp%20Form) for $SL_2(\mathbb{Z})$, the translation matrix gives $f(z+1)=f(z)$ and $g(z+1)=g(z)$. This periodicity means that, near the cusp, both functions descend through the local coordinate $q=e^{2\pi iz}$ to holomorphic functions on a punctured disc around $0$. The cusp condition says more: those descended functions extend holomorphically to $q=0$ and have value $0$ there. Therefore, after factoring out the first power of $q$, there are $r\in(0,1)$ and holomorphic maps
\begin{align*}
F,G:\{w\in\mathbb{C}:|w|<r\}\to\mathbb{C}
\end{align*}
such that whenever $|q(z)|<r$,
\begin{align*}
f(z)&=q(z)F(q(z)),\\
g(z)&=q(z)G(q(z)).
\end{align*}
This factorisation is the rigorous reason the whole infinite [Fourier series](/page/Fourier%20Series) decays uniformly like the first power of $q$: the remaining factors $F$ and $G$ are bounded on a smaller closed disc. Choose $Y_0\geq 1$ so that $e^{-2\pi Y_0}<r/2$. Since the closed disc $\{w\in\mathbb{C}:|w|\leq e^{-2\pi Y_0}\}$ is compactly contained in the domain of $F$ and $G$, continuity gives finite constants
\begin{align*}
M_f&:=\sup_{|w|\leq e^{-2\pi Y_0}}|F(w)|,\\
M_g&:=\sup_{|w|\leq e^{-2\pi Y_0}}|G(w)|.
\end{align*}
Set $C_f:=M_f$ and $C_g:=M_g$. If $z=x+iy\in\mathcal{F}_{>Y_0}$, then $|q(z)|=e^{-2\pi y}\leq e^{-2\pi Y_0}$, and hence
\begin{align*}
|f(z)|&=|q(z)|\,|F(q(z))|\leq C_f e^{-2\pi y},\\
|g(z)|&=|q(z)|\,|G(q(z))|\leq C_g e^{-2\pi y}.
\end{align*}
Multiplying the estimates gives
\begin{align*}
|f(z)\overline{g(z)}|\,y^{k-2}
\leq
C_fC_g e^{-4\pi y}y^{k-2}.
\end{align*}
The factor $y^{k-2}$ is only polynomial in $y$, whereas $e^{-4\pi y}$ decays exponentially. Therefore
\begin{align*}
\int_{Y_0}^{\infty} C_fC_g e^{-4\pi y}y^{k-2}\,d\mathcal{L}^1(y)<\infty.
\end{align*}
Finally, since the cusp part of the fundamental domain lies inside the strip $[-1/2,1/2]\times(Y_0,\infty)$, enlarging the domain of integration to that strip gives
\begin{align*}
\int_{\mathcal{F}_{>Y_0}} |f(z)\overline{g(z)}|\,y^{k-2}\,d\mathcal{L}^2(z)
\leq
\int_{Y_0}^{\infty} C_fC_g e^{-4\pi y}y^{k-2}\,d\mathcal{L}^1(y)
<\infty.
\end{align*}[/guided]