This technique is sometimes called double-counting, especially when the sums are finite. Assuming that Fubini's theorem is applicable, we have
\begin{align*}
\sum_{n=1}^{\infty}\sum_{i=nr}^{\infty} f(i)
= \sum_{\substack{i, n \geq 1 \\ i \geq nr}} f(i)
= \sum_{\substack{i, n \geq 1 \\ n \leq i/r}} f(i)
= \sum_{i=1}^{\infty} \sum_{n \, : \, 1\leq n \leq i/r} f(i).
\end{align*}
In the last expression, $f(i)$ is independent of $n$. So the value of the sum is simply $f(i)$ times the number of summands, which is exactly $\lfloor i/r \rfloor$, i.e.,
$$ \sum_{n \, : \, n \leq i/r} f(i) = (\#\{n : 1\leq n \leq i/r \}) f(i) = \left\lfloor\frac{i}{r}\right\rfloor f(i). $$
Therefore the desired identity follows.
Here is a way of visualizing the situation. For the sum $\sum_{n=1}^{\infty} \sum_{i=nr}^{\infty} a_{n,i}$, we represent the term $a_{n,i}$ as the dot at $(n, i)$. Then the range of this double sum can be visualized as:

The above figure corresponds to $r = \sqrt{2}$. Then $\sum_{n=1}^{\infty} \sum_{i=nr}^{\infty} a_{n,i}$ amounts to summing over each column first. If we sum over each row first, then the bound of $n$ for the '$i$-th row' is determined by the inequality $i \geq nr$, or equivalently, $n \leq i/r$. Since $n$ only takes integer values, this is equivalent to $n \leq \lfloor i/r \rfloor$, hence the identity
$$ \sum_{n=1}^{\infty} \sum_{i=nr}^{\infty} a_{n,i} = \sum_{i=1}^{\infty} \sum_{n=1}^{\lfloor i/r \rfloor} a_{n,i} $$
follows under suitable condition on $a_{n,i}$.