I won't reproduce the proof here, but here is a link to another MathSE question about the same theorem. I've read Brian M. Scott's answer, but I think I'm missing something — I still don't understand why $m_1$ and $m_2$ are necessary for the proof; why can't we just pick an $m$ like in the following examples?
$$ \text{Let } m = \begin{cases} 5 &\text{if } nx = 4.6 \\ 2 &\text{if } nx = \sqrt{2} \\ -6 &\text{if } nx = -7 \\ \vdots &\text{etc.} \end{cases} , $$
i.e. let $m$ be the next largest integer. I believe this method will always satisfy $m - 1 \leq nx < m$.
Now I'm guessing that the reason we can't directly pick an $m$ without the help of $m_1$ and $m_2$ has something to do with rigour; I think I would be assuming things about $\mathbb{R}$ that Rudin wouldn't want me to just yet, but I can't quite put my finger on what exactly.