[guided]We want to compute the derivative of
\begin{align*}
F(t)=\int_M u(x,t)v(x,t)\,d\mu_{g(t)}(x),
\end{align*}
but there are two sources of time dependence: the functions $u$ and $v$, and the measure $d\mu_{g(t)}$. Since $M$ is compact and all objects are smooth, differentiating under the integral sign is valid. The product rule gives
\begin{align*}
\partial_t(uv)(x,t)=(\partial_t u)(x,t)v(x,t)+u(x,t)(\partial_t v)(x,t).
\end{align*}
We also need the time derivative of the Riemannian measure. Let $(U_{\mathrm{chart}},\varphi)$ be a coordinate chart on $M$, where $U_{\mathrm{chart}}\subset M$ is the chart domain and $\varphi:U_{\mathrm{chart}}\to\varphi(U_{\mathrm{chart}})\subset\mathbb{R}^n$ is the coordinate map. In this chart, let $g_{ij}(x,t)$ be the components of $g(t)$ and define the local density $\theta: \varphi(U_{\mathrm{chart}})\times I\to (0,\infty)$ by
\begin{align*}
\theta(x,t):=\sqrt{\det(g_{ij}(x,t))},
\end{align*}
where $x\in\varphi(U_{\mathrm{chart}})$ is the coordinate variable. Let $\mathcal{L}^n$ denote $n$-dimensional Lebesgue measure on $\mathbb{R}^n$. Locally, the Riemannian measure is represented by $d\mu_{g(t)}=\theta(x,t)\,d\mathcal{L}^n(x)$. The determinant differential identity gives
\begin{align*}
\partial_t\theta=\frac{1}{2}\operatorname{tr}_{g(t)}(\partial_t g(t))\theta.
\end{align*}
Since $g(t)$ solves Ricci flow, $\partial_tg(t)=-2\operatorname{Ric}_{g(t)}$. Taking the $g(t)$-trace and using $R_{g(t)}=\operatorname{tr}_{g(t)}\operatorname{Ric}_{g(t)}$ gives
\begin{align*}
\partial_t\theta=-R_{g(t)}\theta.
\end{align*}
This local density identity is invariant under coordinate changes, hence
\begin{align*}
\partial_t d\mu_{g(t)} = -R_{g(t)}\,d\mu_{g(t)}.
\end{align*}
Therefore
\begin{align*}
F'(t)=\int_M \partial_t(uv)(x,t)\,d\mu_{g(t)}(x)+\int_M u(x,t)v(x,t)\,\partial_t d\mu_{g(t)}(x).
\end{align*}
Using the product rule and the volume variation identity, this becomes
\begin{align*}
F'(t)=\int_M \left((\partial_t u)(x,t)v(x,t)+u(x,t)(\partial_t v)(x,t)\right)\,d\mu_{g(t)}(x)-\int_M R_{g(t)}(x)u(x,t)v(x,t)\,d\mu_{g(t)}(x).
\end{align*}
Now we use the two PDEs. The forward heat equation gives
\begin{align*}
\partial_t u = \Delta_{g(t)}u,
\end{align*}
and the conjugate heat equation gives
\begin{align*}
\partial_t v = -\Delta_{g(t)}v+R_{g(t)}v.
\end{align*}
Substituting these identities into the differentiated pairing yields
\begin{align*}
F'(t)=\int_M v(x,t)\Delta_{g(t)}u(x,t)\,d\mu_{g(t)}(x)+\int_M u(x,t)\left(-\Delta_{g(t)}v(x,t)+R_{g(t)}(x)v(x,t)\right)\,d\mu_{g(t)}(x)-\int_M R_{g(t)}(x)u(x,t)v(x,t)\,d\mu_{g(t)}(x).
\end{align*}
The scalar curvature term from the conjugate heat equation is
\begin{align*}
\int_M R_{g(t)}(x)u(x,t)v(x,t)\,d\mu_{g(t)}(x),
\end{align*}
and the scalar curvature term from the evolving measure is its negative. These two terms cancel exactly. Hence
\begin{align*}
F'(t)
= \int_M v(x,t)\Delta_{g(t)}u(x,t)\,d\mu_{g(t)}(x)
- \int_M u(x,t)\Delta_{g(t)}v(x,t)\,d\mu_{g(t)}(x).
\end{align*}
This is the point of the conjugate heat equation: the additional $R_{g(t)}v$ term compensates precisely for the loss of fixed volume under Ricci flow.[/guided]