The answer is no, i.e. the solution operator of the boundary problem value of every non-homogeneous linear equation is not a semigroup of operators respect to any of the space or time variables involved. I will show this in two steps: I'll construct the general solution operator $T$ for the Dirichlet problem for the Poisson equation in the upper half plane and then show that this cannot be a semigroup.
First Step. Let's precisely define the Green function for a non homogeneous boundary value problem and explicitly calculate it for the Dirichlet problem in the upper half-plane for the Poisson equation as required by the OP, i.e.
$$
\begin{cases}
\Delta u(x,y)=f(x,y),\quad (x,y)\in\mathbb{R}\times\mathbb{R}^+\\
u(x,0)=g(x).
\end{cases}\tag{1}\label{1}
$$
Definition. The Green function of the (linear) boundary value problem
$$
\begin{cases}
P(x,D) u(x)=f(x),\quad x\in G\subset\mathbb{R}^n\\
Bu(x)=g(x),\quad x\in\partial G
\end{cases}\tag{2}\label{2}
$$
for the linear partial differential operator $P(x,D)$ is the distribution solution of the following associated boundary problem
$$
\begin{cases}
P(x,D) \mathscr{G}(x,t)=\delta(x-t),\quad x,t\in G\subset\mathbb{R}^n\\
B\mathscr{G}(x,t)=0,\quad x\in\partial G
\end{cases}\tag{3}\label{3}
$$
where
- $G$ is a domain with a "sufficiently regular" boundary $\partial G$,
- $B$ is a linear boundary operator defined on $\partial G$.
Notes.
- Clearly the Dirichlet problem \eqref{1} is of type \eqref{2} since $B u(x,y)=u|_{\partial G}=u(x,0)$ is a linear boundary operator, and
- a solution of problem \eqref{3} for \eqref{1} exists under reasonable conditions on $\partial G$ and on $P(x,D)$ (Vladimirov (1983) §29.1, p. 369).
The Green's function $\mathscr{G}((x,y),(t,s))$ for \eqref{1} is the solution of the following Dirichlet boundary value problem:
$$
\begin{cases}
\Delta \mathscr{G}((x,y),(t,s))=\delta((x,y)-(t,s)),\quad (x,y),(t,s)\in \mathbb{R}\times\mathbb{R}^+\\
\mathscr{G}((x,0),(t,s))=0.
\end{cases}\tag{3'}\label{3'}
$$
Note. The vector $(t,s)\in \mathbb{R}\times\mathbb{R}^+$ is a parameter which has an interesting physical interpretation in the theory of the static electric field (see Vladimirov (1983) §29.1 for the details)
The solution of problem \eqref{3'} has the form (Vladimirov (1983) §29.1, p. 368)
$$
\mathscr{G}((x,y),(t,s))=\frac{1}{2\pi}\ln|(x,y)-(t,s)|+g((x,y),(t,s))
$$
where the first term on the right side is the fundamental solution of the laplacian while the second one is a function harmonic in the whole $\mathbb{R}\times\mathbb{R}^+$. In the problem posed by the OP,
$$
\mathscr{G}((x,y),(t,s))=\frac{1}{2\pi}\big(\ln|(x,y)-(t,s)|-\ln|(x,y)-(t,-s)|\big).\tag{4}\label{4}
$$
Now, noting that $\frac{\partial\mathscr{G}}{\partial\mathbf{n}}=\frac{\partial\mathscr{G}}{\partial t}$ on the boundary $\{y=0\}$ of the upper half plane, by Green's formula we obtain the formula for the general solution of \eqref{1}:
$$
\begin{split}
u(x,y)&=\int\limits_{\mathbb{R}\times\mathbb{R}^+}\mathscr{G}((x,y),(t,s))f(t,s)\mathrm{d}t\mathrm{d}s+\int\limits_{\{s=0\}}g(t)\frac{\partial\mathscr{G}}{\partial t}((x,y),(t,0))\mathrm{d}t\\
&=\int\limits_{\mathbb{R}\times\mathbb{R}^+}\mathscr{G}((x,y),(t,s))f(t,s)\mathrm{d}t\mathrm{d}s+\frac{1}{\pi}\int\limits_{-\infty}^\infty \frac{y}{y^2+(x-t)^2}g(t)\mathrm{d}t\\
&\overset{\mathrm{def}}{=} T(g;f)(x,y)
\end{split}\tag{5}\label{5}
$$
Second step. Now note that the operator $T$ defined by \eqref{5} is a semigroup only if $f\equiv 0$. To see this, suppose that we have homogeneous boundary conditions, i.e. $g\equiv 0$ and assume $y$ as the parameter of the hypothetical semigroup, i.e $T(g;f)(x,y)=T_y(g;f)(x)$: for $y=0$ we have
$$
T_0(0;f)(x)\equiv 0 \quad \forall x\in\mathbb{R}\text{ by equation \eqref{4} }
$$
If $T_y$ defined by \eqref{5} would be a semigroup, the equation above would imply that
$$
T_y(0,f)(x)=T_{y+0}(0,f)=T_{y}T_{0}(0,f)(x)\equiv 0
$$
by the second property of the semigroup and by the linearity of $T$, and this is clearly false since the first term of \eqref{5} is not necessarily $\equiv0$. If instead of $y$ we try to assume $x$ as the parameter of the hypothetical semigroup by posing $T(g;f)(x,y)=T_x(g;f)(y)$, it is simple to see that $T_0\neq \mathrm{id}$.
Last notes
- This proposition holds for general boundary problems \eqref{2} since it can be proved that, for the general Green's function defined by \eqref{3}
$$
B\mathscr{G}(x,t)=0,\quad t\in\partial G.
$$
See again Vladimirov (1983) §29.1 for the details when $P(x,D)=\Delta$.
- Alberto Cialdea showed me the main argument of the second step in the proof above, and I would like to thank him publicly.
[1] Vladimirov, V. S. (1983)[1970], Equations of mathematical physics, Moscow: Mir Publishers, 2nd ed., pp. 464, MR0764399, Zbl 0207.09101
(the Zbmath review refers to the first English edition).