$f(x)$ is a value, not a function, so technically it's wrong to say that $f(x)$ is differentiable. To be precise you should say "$h'(x) = f'(x) + g'(x)$ for any $x \in D$ for any differentiable functions $f,g$ where $h(x) = f(x) + g(x)$ for any $x \in D$", but it is more concise and equally precise to say "$(f+g)' = f' + g'$ on D". Likewise the middle line with $\frac{d}{dx} f$ is probably the 'original' Leibniz notation, but it makes no sense if $f,g,h$ are strictly functions; the only way to get the correct semantic meaning is to interpret $f,g,h$ not as functions but as values that are determined by $x$. Historically $\frac{df}{dx}$ meant the change in $f$ divided by the change in $x$, with "d" standing for "Δ" / "δ" (delta). But if you want to interpret $f,g,h$ as true (1-variable) functions, then none of them are externally bound to any variables, so $x$ has no relation to $f,g,h$.
At the same time, Wikipedia says that Leibniz's notation is indeed in the form of the last line, but again we cannot interpret $f(x)$ within the $\frac{d}{dx}$ as the value of $f$ on input $x$, but rather have to interpret $\frac{d}{dx}(f(x))$ as something more like $\frac{d}{dx}($"f(x)"$)$ where $\frac{d}{dx}$ behaves like a procedure that looks at how the value described by the input string containing some occurrences of "$x$" depends on the value of $x$. This somewhat strange abuse of notation is quite common in many areas of mathematics, which you should be aware of, and when using such notation you should be clear about which expressions denote the objects themselves, which denote their descriptions, and which denote their state as bound to the state of other objects.