[step:Concatenate the old and new solutions to contradict maximality]
Define an interval
\begin{align*}
\widetilde{J}:=J\cup[b,b+\varepsilon].
\end{align*}
We verify that this union has no gap. If $q\in J$ and $q<r<b$, then $r\in J$ because $J$ is an interval and there exists $p\in J$ with $r<p<b$, using $b=\sup J$. Hence $J$ contains all points of $I$ lying between any sufficiently late point of $J$ and $b$, so adjoining $[b,b+\varepsilon]$ produces an interval. Since $[b,b+\varepsilon]\subset I$, we have $\widetilde{J}\subset I$, and because $b+\varepsilon\in\widetilde{J}$ while $b+\varepsilon\notin J$, the containment $J\subsetneq\widetilde{J}$ is proper.
Define
\begin{align*}
\widetilde{x}:\widetilde{J}\to U
\end{align*}
by
\begin{align*}
\widetilde{x}(t)=x(t)\quad\text{for }t\in J,\ t<b,
\end{align*}
and
\begin{align*}
\widetilde{x}(t)=z(t)\quad\text{for }t\in[b,b+\varepsilon].
\end{align*}
If $b\in J$, then the previously established limit gives $x(b)=y=z(b)$, because $x$ is continuous. If $b\notin J$, the same equality gives the continuous value inserted at the joining time. Thus $\widetilde{x}$ is continuous at $b$ and agrees with a solution on each side of $b$.
For every compact subinterval of $\widetilde{J}$ not crossing $b$, the differential equation holds because either $\widetilde{x}=x$ or $\widetilde{x}=z$. At $b$, the right derivative of $\widetilde{x}$ is the right derivative of $z$, equal to $F(b,y)$, and the left derivative, when required by the endpoint convention for solutions, follows from the left-hand equation for $x$ together with continuity of $F$ and $x(t)\to y$. Hence $\widetilde{x}$ is a solution on $\widetilde{J}$.
This solution extends $x$ to a strictly larger interval contained in $I$, contradicting maximality of $x:J\to U$. Therefore the assumed compact set $C$ and number $a<b$ cannot exist. Equivalently, for every compact set $C\subset U$ and every $a<b$, there exists $t\in J$ such that $a<t<b$ and $x(t)\notin C$.
[/step]