Proof
First, we may prove $$x_n^2 \geq 2n-1,(n=1,2,\cdots).\tag1$$ Obviously, $(1)$ holds for $n=1$. Assume that $(1)$ holds for $n=k$, then $$x_{k+1}^2=x_k^2+2+\frac{1}{x_k^2}\geq x_k^2+2\geq 2n-1+2=2(n+1)-1,$$ which implies $(1)$ also holds for $n=k+1.$ Thus, by induction, $(1)$ holds for all $n=1,2,\cdots.$
Therefore, $\forall M>0$, $\exists N=[\dfrac{M^2+1}{2}]+1$ such that $$x_n \geq \sqrt{2n-1}>M$$ holds for $n>N$. We are done.
Your strategy is valid. However, you needn't have spotted a function $f$ for which $f(x_{n+1})-f(x_n)$ has a positive lower bound. Note that the sequence is strictly increasing, so to diverge it just needs to not have a finite limit $L$. And if such an $L$ existed we'd have $L=L+L^{-1}$, which no finite $L$ satisfies. In other words, $x_{n+1}=x_n+g(x_n)$ with $x\ge x_1\implies g(x)>0$ will always diverge to $\infty$.