I don't really get why testhalt doesnt exist based on the proof that some program, ie turing, called on testhalt does the opposite of testhalt.
Below is a definition for testhalt
And one for turing.
How come we absolutely must, in turing, do the opposite of testhalt? (meaning loop if testhalt halts, for example)
The point isn't that $\mathsf{TURING}$ does the opposite of $\mathsf{TESTHALT}$, but rather that - when fed a judiciously chosen input - it does the opposite of itself. Since this is blatantly impossible, $\mathsf{TURING}$ can't exist and therefore $\mathsf{TESTHALT}$ can't exist either.
Specifically, consider what happens when we feed $\mathsf{TURING}$ its own index $\tau$. We get that $\mathsf{TURING}(\tau)$ halts iff $\mathsf{TESTHALT}(\tau,\tau)$ says "no" (by definition of $\mathsf{TURING}$) iff $\mathsf{TURING}$ does not halt on input $\tau$ (by assumption on $\mathsf{TESTHALT}$). This self-contradictory behavior ("I halt on input $\tau$ ifff I don't halt on input $\tau$") is what concludes the argument.