[guided]We need to prove that the product has the same symbol estimates, with the product order function $m_1m_2:T^*\mathbb{R}^n \to (0,\infty)$ defined by
\begin{align*}
(m_1m_2)(x,\xi)=m_1(x,\xi)m_2(x,\xi).
\end{align*}
Define the product symbol
$p:T^*\mathbb{R}^n \times (0,1] \to \mathbb{C}$ by
\begin{align*}
p(x,\xi;h)=a(x,\xi;h)b(x,\xi;h).
\end{align*}
To show $p\in S_\delta(m_1m_2)$, we must estimate every mixed derivative $\partial_x^\alpha\partial_\xi^\beta p$. Fix multi-indices $\alpha,\beta \in \mathbb{N}_0^n$. The correct tool is the multi-index Leibniz formula, because a derivative of a product is a finite sum over all ways to distribute the $x$-derivatives and $\xi$-derivatives between the two factors:
\begin{align*}
\partial_x^\alpha \partial_\xi^\beta p(x,\xi;h)=\sum_{\mu\leq\alpha}\sum_{\nu\leq\beta}\binom{\alpha}{\mu}\binom{\beta}{\nu}(\partial_x^\mu\partial_\xi^\nu a)(x,\xi;h)(\partial_x^{\alpha-\mu}\partial_\xi^{\beta-\nu} b)(x,\xi;h).
\end{align*}
Here $\mu\leq\alpha$ means $\mu_i\leq\alpha_i$ for every component $i\in\{1,\dots,n\}$, and similarly $\nu\leq\beta$. The sums are finite because there are only finitely many such multi-indices.
Now we estimate one summand. Since $a\in S_\delta(m_1)$, the defining estimate gives a constant $A_{\mu,\nu}>0$ such that
\begin{align*}
|\partial_x^\mu\partial_\xi^\nu a(x,\xi;h)| \leq A_{\mu,\nu} h^{-\delta(|\mu|+|\nu|)}m_1(x,\xi).
\end{align*}
Since $b\in S_\delta(m_2)$, the defining estimate gives a constant $B_{\alpha-\mu,\beta-\nu}>0$ such that
\begin{align*}
|\partial_x^{\alpha-\mu}\partial_\xi^{\beta-\nu} b(x,\xi;h)| \leq B_{\alpha-\mu,\beta-\nu} h^{-\delta(|\alpha-\mu|+|\beta-\nu|)}m_2(x,\xi).
\end{align*}
Multiplying the two inequalities gives
\begin{align*}
|(\partial_x^\mu\partial_\xi^\nu a)(x,\xi;h)(\partial_x^{\alpha-\mu}\partial_\xi^{\beta-\nu} b)(x,\xi;h)| \leq A_{\mu,\nu}B_{\alpha-\mu,\beta-\nu} h^{-\delta(|\mu|+|\nu|+|\alpha-\mu|+|\beta-\nu|)}m_1(x,\xi)m_2(x,\xi).
\end{align*}
The exponent is exactly the exponent required for an $(\alpha,\beta)$ derivative, because multi-index length is additive under this splitting:
\begin{align*}
|\mu|+|\alpha-\mu|+|\nu|+|\beta-\nu|=|\alpha|+|\beta|.
\end{align*}
Thus each summand is bounded by a constant times
\begin{align*}
h^{-\delta(|\alpha|+|\beta|)}m_1(x,\xi)m_2(x,\xi).
\end{align*}
Because the Leibniz formula contains only finitely many summands, we may collect their constants into one finite constant:
\begin{align*}
C_{\alpha,\beta}=\sum_{\mu\leq\alpha}\sum_{\nu\leq\beta}\binom{\alpha}{\mu}\binom{\beta}{\nu}A_{\mu,\nu}B_{\alpha-\mu,\beta-\nu}.
\end{align*}
Taking absolute values in the Leibniz formula and applying the triangle inequality gives
\begin{align*}
|\partial_x^\alpha \partial_\xi^\beta p(x,\xi;h)| \leq C_{\alpha,\beta}h^{-\delta(|\alpha|+|\beta|)}m_1(x,\xi)m_2(x,\xi).
\end{align*}
This is precisely the defining estimate for $p\in S_\delta(m_1m_2)$. Since $\alpha$ and $\beta$ were arbitrary, the product $ab$ belongs to $S_\delta(m_1m_2)$.[/guided]