There are several approaches, though fundamentally the idea is the same: when a series converges, it can be made bigger while preserving convergence.
Consider the sequence $$f_n(x) = \begin{cases} |f(x)|, \quad &|f(x)|\ge n, \\ 0 \quad & |f(x)|<n \end{cases} \tag{1}$$
Observe that $\|f_n\|_{L^p}\to 0$ as $n\to\infty$. Therefore, there is a stricly increasing sequence $n_k$ such that $\|f_{n_k}\|_{L^p}\le 2^{-k}$. Define $g(x) = \sum_k f_{n_k}$, which is in $L^p$.
By construction, $g$ is of the form $\psi(|f|)$ with $$\psi = \sum_{k} \chi_{[n_k,\infty)} $$
Hence, $\psi(s)/s\to\infty $ as $s\to\infty$.
However there is the annoying requirement of continuity. One way around it is to replace (1) with
$$f_n(x) = \begin{cases} |f(x)|, \quad &|f(x)|\ge n+1, \\ |f(x)|\left(|f(x)|-n\right), \quad &n\le |f(x)|\le n+1, \\ 0 \quad & |f(x)|<n \end{cases} \tag{2}$$
and repeat. The function $\psi$ will now be the sum of "tapered" characteristic functions. Continuity of the sum is easy to check because only one summand changes at a time.