Every symmetric, positive-definite matrix has a square root.
In particular, the root may be asked to be symmetric and positive-definite as well, and then it is uniquely determined.
For the given discretisation matrix $A^h$, which is highly structured, (t-)his square root
$$\sqrt{A^h} \;=\; \frac1{h^2}
\begin{pmatrix}
2 & -1 & 0 & &\\ -1 & 2 & -1 & & & \\
0 & -1 & \ddots & \ddots &\\ & & \ddots & \ddots & -1 & 0\\
& & & -1 & 2 & -1\\ & & & 0 & -1 & 2
\end{pmatrix}.$$
is not so far away, let's say by an educated guess which extends
$$\begin{pmatrix}
5 & -4\\ -4 & 5
\end{pmatrix} \;=\;
\begin{pmatrix}
2 & -1\\ -1 & 2
\end{pmatrix}^2\:.$$
And $\sqrt{A^h}$ is positive-definite:
Cf$\,$
https://en.wikipedia.org/wiki/Definite_matrix#Examples for the $3\times 3$ matrix. The method should generalise to higher dimensions.
And see here on this site:
The full statement including proof of $(7.4.7)$ Theorem starts on page 537 in https://zhilin.math.ncsu.edu/TEACHING/MA580/Stoer_Bulirsch.pdf .
An explicit Sum of squares expression as looked for in the OP is given by
$$(2u_1-u_2)^2 \,+\, \sum_{j=2}^{n-1}\,(-u_{j-1}+2u_j-u_{j+1})^2 \,+\, (2u_n-u_{n-1})^2\tag{SoS}$$
Having the knowledge of the square root $\sqrt{A^h}\,$ it results from expanding
\begin{align}
h^4\cdot\mathbf{u}^TA^h\,\mathbf{u} & \:=\:
\left(h^2\cdot\mathbf{u}^T\sqrt{A^h}\right) \left(h^2\cdot\sqrt{A^h}\,\mathbf{u}\right) \\[2ex]
& \:=\:
\big(2u_1-u_2, \dots, -u_{n-1}+2u_n\big)
\begin{pmatrix}
2u_1 -u_2\\ -u_1 +2u_2 -u_3\\
\vdots \\
-u_{n-2} +2u_{n-1} -u_n \\ -u_{n-1} +2u_n
\end{pmatrix}
\end{align}
The expression $(\text{SoS})$ gets zero only if $\mathbf{u}= \mathbf{0}$.
This is equivalent to the linear system $\sqrt{A^h}\,\mathbf{u} = \mathbf{0}$ having only the trivial solution.
Best Answer
You can write $A$ as the sum of rank 1 matrices: $$ \begin{pmatrix}a_1 & \dots & a_1 \\ \vdots & \ddots \\ a_1& & a_1 \end{pmatrix} + \begin{pmatrix}0 & 0 & \dots & 0 \\ 0 & a_2 - a_1 & \dots & a_2-a_1 \\ \vdots & \vdots & \ddots \\ 0 & a_2 - a_1& & a_2 - a_1 \end{pmatrix} + \text{etc} $$ The $i^{th}$ term has eigenvector $(0, 0, \ldots, 0, 1, 1, \ldots 1)$, where the number of zeros is $i-1$ and eigenvalue $(n+1-i)(a_i - a_{i-i}) \geq 0$ (with $a_0=0$). Since the matrices have rank $1$, the remaining eigenvalues are $0$.
The sum of positive semidefinite matrices is positive semidefinite.