Given $\varepsilon>0$, one can find $N$ so large that whenever $n > N$ then $|a_n -L| < \dfrac \varepsilon 2$.
Then
\begin{align}
& \frac{a_1 + \cdots + a_N + a_{N+1} + \cdots + a_M} M \\[10pt]
= {} & \frac N M \cdot \underbrace{\frac{a_1+\cdots+a_N} N}_\text{first average} {}+{} \frac{M-N} M \cdot \underbrace{\frac{a_{N+1} + \cdots + a_M}{M-N}}_\text{second average}.
\end{align}
The second average above is the average of numbers between $L\pm\dfrac\varepsilon2$, and is thus itself between those bounds.
As $M$ grows while $N$ stays fixed, we have $\dfrac N M\to 0$ so the whole term involving the first average approaches $0$, and $\dfrac{M-N} M \to 1$.
For large enough $M$, see if you can show the whole thing must therefore be between $L\pm\varepsilon$.