Objective:
The objective is to compute the quantity:
$$
\langle n|(a+a^{\dagger})^k|m\rangle
$$
the Fock states, $|m\rangle$, normalised such that $\langle n|m\rangle=\delta_{n,m}$, and $[a,a^\dagger]=1$.
Preliminary Result:
It is convenient to use coherent state techniques, so in particular we primarily need the relation:
\begin{equation}
\boxed{
\begin{aligned}
I(z,w)&\equiv\langle \bar{z} | (a+a^\dagger)^k |w\rangle\\
& = B_k(w+z,1,0,\dots,0) \,e^{wz}
\end{aligned}
}\qquad (\star)
\end{equation}
where $\langle \bar{z}|=\langle 0|e^{za}$ and $|w\rangle=e^{wa^\dagger}|0\rangle$ are coherent states (with $a|w\rangle = w|w\rangle$ and $\langle\bar{z}|a^\dagger=z\langle\bar{z}|$, and $\langle \bar{z}|w\rangle = e^{wz}$), whereas $B_k(w+z,1,0,\dots,0)$ is a complete Bell polynomial. As usual, we normalise $[a,a^\dagger]=1$.
Derivation of ($\star$):
Since I received some questions about how to derive $(\star)$ let me point out that I made use of the complete Bell polynomial identity, $B_k(a_1,\dots,a_n) = \partial_t^k \exp(\sum_{s=1}^\infty \frac{1}{s!}a_st^s)|_{t=0}$, and the Baker-Campbell-Hausdorff formula, which reduces to $e^{X+Y}=e^X e^Y e^{-\frac{1}{2}[X,Y]}$, when $[X,Y]$ commutes with $X$ and $Y$; from the latter (since the right-hand side must be symmetric in $X,Y$ as is the left-hand side) it also follows that $e^X e^Y = e^Ye^Xe^{[X,Y]}$. In further detail, making use of these results,
\begin{equation}
\begin{aligned}
I(z,w) &\equiv\langle \bar{z} | (a+a^\dagger)^k |w\rangle\\
&=\partial_t^k\langle \bar{z} | e^{(a+a^\dagger)t} |w\rangle\big|_{t=0}\\
&=\partial_t^k\langle \bar{z} | e^{a^\dagger t} e^{at}e^{\frac{1}{2}[a,a^\dagger]t^2} |w\rangle\big|_{t=0}\\
&=\partial_t^ke^{(z+w)t+\frac{1}{2}t^2}\big|_{t=0} \langle \bar{z} |w\rangle\\
&=B_k(w+z,1,0,\dots,0) \,e^{wz}
\end{aligned}
\end{equation}
Main Calculation:
Returning to ($\star$), we can extract the quantity of interest from it by differentiating it wrt $z$ and $w$ ($n$ and $m$ times respectively) and then setting $z=w=0$; after including relevant normalisations (assuming $\langle n|m\rangle=\delta_{n,m}$):
$$
\boxed{
\begin{aligned}
\langle n|(a+a^{\dagger})^k|m\rangle
&=\frac{1}{\sqrt{n!m!}}\partial_z^n\partial_w^mI(z,w)\big|_{z,w=0}\\
&=\frac{1}{\sqrt{n!m!}}\partial_z^n\partial_w^m\,B_k(w+z,1,0,\dots,0) \,e^{wz}\big|_{z=w=0}\\
\end{aligned}
}\qquad (\star\star)
$$
To evaluate the derivatives, I find it convenient to work in terms of cycle index polynomials, $Z_k(a_s)\equiv Z_k(a_1,a_2,\dots,a_k)$, defined by:
$$
Z_k(a_1,a_2,\dots,a_k) = \frac{1}{k!}B_k(0!a_1,1!a_2,\dots,(k-1)!a_k).
$$
All the properties of cycle index polynomials we need are given in Appendix B of the long paper. In particular, we need the following result:
\begin{equation}
\begin{aligned}
\partial_z^aZ_k(z+w,1,0,\dots,0)&=Z_{k-a}(z+w,1,0,\dots,0),
\end{aligned}
\end{equation}
which follows immediately from the derivative relation in the long paper, and also:
$$
Z_{2p}(0,1,0,\dots,0)=\frac{2^{-p}}{p!},\quad
Z_{2p+1}(0,1,0,\dots,0) = 0,\qquad(\star\star\star)
$$
for some $p=0,1,2,\dots$, which follow immediately, e.g., from the contour integral representation in equation (B.677a) in the long paper.
Plugging these into the above, some elementary manipulations then lead to:
\begin{equation}
\begin{aligned}
\langle n|(a+a^{\dagger})^k|m\rangle
&=\frac{k!}{\sqrt{n!m!}}\partial_z^n\partial_w^m\,Z_k(w+z,1,0,\dots,0) \,e^{wz}\big|_{z=w=0}\\
&=\frac{k!}{\sqrt{n!m!}}\partial_z^n\sum_{b=0}^m\binom{m}{b}\big(\partial_w^b Z_k(w+z,1,0,\dots,0)\big)\big(\partial_w^{m-b}e^{wz}\big)\big|_{z=w=0}\\
&=\frac{k!}{\sqrt{n!m!}}\partial_z^n\sum_{b=0}^m\binom{m}{b} Z_{k-b}(z,1,0,\dots,0)z^{m-b}\big|_{z=0}\\
\end{aligned}
\end{equation}
where we just used the usual Leibniz rule for differentiating a product $m$ times. Similarly, carrying out the remaining derivatives,
\begin{equation}
\begin{aligned}
\langle n|(a+a^{\dagger})^k|m\rangle
&=\frac{k!}{\sqrt{n!m!}}\sum_{a=0}^n\binom{n}{a}\sum_{b=0}^m\binom{m}{b} \big(\partial_z^aZ_{k-b}(z,1,0,\dots,0)\big)\big(\partial_z^{n-a}z^{m-b}\big)\big|_{z=0}\\
&=\frac{k!}{\sqrt{n!m!}}\sum_{a=0}^n\binom{n}{a}\sum_{b=0}^m\binom{m}{b} Z_{k-a-b}(0,1,0,\dots,0)(m-b)!\delta_{n-a,m-b}\\
&=\sum_{a=0}^n\frac{k!\sqrt{n!m!}}{(n-a)!a!(m-n+a)!} Z_{k-m+n-2a}(0,1,0,\dots,0)\\
\end{aligned}
\end{equation}
where in the last line we used the the Kronecker delta to carry out the sum over $b$. From $(\star\star\star)$ and the above relation, it is clear that unless a positive integer, $r$, can be found such that,
$$
k=2r+m-n,
$$
the quantity $\langle n|(a+a^\dagger)^k|m\rangle$ will vanish. We can then carry out the sum over $a$, taking into account $(\star\star\star)$, to arrive at:
$$
\begin{aligned}
\langle n|(a+a^{\dagger})^k|m\rangle
&=\sum_{a=0}^n\frac{(2r+m-n)!\sqrt{n!m!}}{(n-a)!a!(m-n+a)!} \frac{2^{-r+a}}{(r-a)!}\\
\end{aligned}
$$
We can now sum over $a$ (I used Mathematica for this sum), we arrive at the final answer:
$$
\boxed{
\langle n|(a+a^{\dagger})^{Q+m-n}|m\rangle = \left\{
\begin{aligned}
&\frac{\sqrt{m!n!}}{2^{Q/2}(m-n)!n!(Q/2)!}\,\,\,\,F_{\!\!\!\!\!\!2\,\,\,\,\,1}\big(-n,-\tfrac{Q}{2};1+m-n;2\big),&\textrm{if}\,\, Q\in 2\mathbf{Z}^+\\
&0&\textrm{if}\,\, Q\in2\mathbf{Z}^++1\\
\end{aligned}
\right.
}
$$
where $m\geq n$.
Consistency Check:
As a consistency check, notice that if $m=n+1$ and $Q=0$ this reduces to the expected answer:
$$
\langle n|(a+a^{\dagger})|n+1\rangle = \sqrt{n+1},
$$
which can alternatively be computed using the commutator, $[a,a^\dagger]=1$, and the Fock representation of the state, $|n\rangle = \frac{1}{\sqrt{n!}}(a^\dagger)^n|0\rangle$.
References:
D. Skliros & D. Luest, Handle Operators in String Theory, Appendix B
Acknowledgement:
Thanks to @Codename47 for pointing out that there was an error in a previous result.