We definitely care that formal logic has a formal semantics. Here's a list of reasons why. More simply, we need to know very precisely what a formalism means. Without this precise definition, there's a lot of wasted time arguing about what things mean.
To answer your other questions, consider the simplest, most accessible case: propositional logic. We have atomic propositions, represented by simple symbols like P and Q, and then formulas built up from the usual connectives like and ($\wedge$), or ($\vee$), and not ($\neg$). Together with the proof theory which governs how we do deductions, we now have a formal system by your definition.
The semantics is built by introducing a separate meta-level, totally outside everything in the previous paragraph. In that level we define two truth values, true and false. In this level we also define a function which maps each propositional symbol to a truth value. Let's call this function I. Here's one possible version of I: {(P,true),(Q,false)}. I is the interpretation function. Finally, we define the usual semantics of the connectives: for example, and ($\wedge$) means both its arguments must be true.
Voila! We have now sketchily defined the semantics of propositional logic. If it sounds like truth tables, it is: each row of a truth table corresponds to one possible interpretation function. Here's a nice writeup.
The nice thing is that all modern formal semantic theories look like the propositional case, just higher-powered: they all have a separate meta-level to define the semantics and an interpretation function to map from the object layer of formulas to the meta-level containing the semantics (roughly speaking).
So now we can quickly answer your questions: a logical system has "semantics/interpretation" because we need to define a separate meta-level and an interpretation function to let us interpret the meanings of well-formed formulas.
A formal system and a logical system are not the same because we can define a formal system that has no formal semantics, say, a context-free grammar or a well-defined card game. Finally, there is additional structure to define the semantics of a logical system, but on a totally separate meta-level.
So you want to prove the following theorem:
Theorem: If $\Gamma,\phi \vdash \psi$ and $\Gamma, \phi \vdash \neg \psi$, then $\Gamma \vdash \neg \phi$
Proof:
First, I'll assume that you can use the Deduction Theorem, which states that for any $\Gamma$, $\varphi$, and $\psi$:
If $\Gamma \cup \{ \varphi \} \vdash \psi$, then $\Gamma \vdash \varphi \rightarrow \psi$
So if $\Gamma,\phi \vdash \psi$ and $\Gamma, \phi \vdash \neg \psi$, then by the Deduction Theorem we have $\Gamma \vdash \phi \to \psi$ and $\Gamma \vdash \phi \to \neg \psi$
This means that if can show that $\phi \to \psi, \phi \to \neg \psi \vdash \neg \phi$, then we're there.
This is not easy, but here goes:
First, let's prove: $\phi \to \psi, \psi \to \chi, \phi \vdash \chi$:
\begin{array}{lll}
1&\phi \to \psi & Premise\\
2& \psi \to \chi & Premise\\
3&\phi& Premise\\
4&\psi& MP \ 1,3\\
5&\chi& MP \ 2,4\\
\end{array}
By the Deduction Theorem, this gives us Hypothetical Syllogism (HS): $\phi \to \psi, \psi \to \chi \vdash \phi \to \chi$
Now let's prove the general principle that $\neg \phi \vdash (\phi \to \psi)$:
\begin{array}{lll}
1. &\neg \phi& Premise\\
2. &\neg \phi \to (\neg \psi \to \neg \phi)& Axiom \ 1\\
3. &\neg \psi \to \neg \phi& MP \ 1,2\\
4. &(\neg \psi \to \neg \phi) \to (\phi \to \psi)& Axiom \ 3\\
5. &\phi \to \psi& MP \ 3,4\\
\end{array}
With the Deduction Theorem, this means $\vdash \neg \phi \to (\phi \to \psi)$ (Duns Scotus Law)
Let's use Duns Scotus to show that $\neg \phi \to \phi \vdash \phi$
\begin{array}{lll}
1. &\neg \phi \to \phi& Premise\\
2. &\neg \phi \to (\phi \to \neg (\neg \phi \to \phi))& Duns \ Scotus\\
3. &(\neg \phi \to (\phi \to \neg (\neg \phi \to \phi))) \to ((\neg \phi \to \phi) \to (\neg \phi \to \neg (\neg \phi \to \phi)))& Axiom \ 2\\
4. &(\neg \phi \to \phi) \to (\neg \phi \to \neg (\neg \phi \to \phi))& MP \ 2,3\\
5. &\neg \phi \to \neg (\neg \phi \to \phi)& MP \ 1,4\\
6. &(\neg \phi \to \neg (\neg \phi \to \phi)) \to ((\neg \phi \to \phi) \to \phi)& Axiom \ 3\\
7. &(\neg \phi \to \phi) \to \phi& MP \ 5,6\\
8. &\phi& MP \ 1,7\\
\end{array}
By the Deduction Theorem, this means $\vdash (\neg \phi \to \phi) \to \phi$ (Law of Clavius)
Using Duns Scotus and the Law of Clavius, we can now show that $ \neg \neg \phi \vdash \phi$:
\begin{array}{lll}
1. &\neg \neg \phi& Premise\\
2. &\neg \neg \phi \to (\neg \phi \to \phi)& Duns \ Scotus\\
3. &\neg \phi \to \phi& MP \ 1,2\\
4. &(\neg \phi \to \phi) \to \phi& Clavius\\
5. &\phi& MP \ 3,4\\
\end{array}
By the Deduction Theorem, this also means that $\vdash \neg \neg \phi \to \phi$ (DN Elim or DNE)
Finally, we can show the desired $\phi \to \psi, \phi \to \neg \psi \vdash \neg \phi$:
\begin{array}{lll}
1. &\phi \to \psi& Premise\\
2. &\phi \to \neg \psi& Premise\\
3. &\neg \neg \phi \to \phi& DNE\\
4. &\neg \neg \phi \to \psi& HS \ 1,3\\
5. &\neg \neg \phi \to \neg \psi& HS \ 2,3\\
6. &(\neg \neg \phi \to \neg \psi) \to (\psi \to \neg \phi)& Axiom \ 3\\
7. &\psi \to \neg \phi& MP \ 5,6\\
8. &\neg \neg \phi \to \neg \phi& HS \ 4,7\\
9. &(\neg \neg \phi \to \neg \phi) \to \neg \phi& Clavius\\
10. &\neg \phi& MP \ 8,9\\
\end{array}
Now, you can actually get a little more quickly to $\neg \neg \phi \vdash \phi$ as follows:
\begin{array}{lll}
1&\neg \neg \phi&Premise\\
2&\neg \neg \phi \to (\neg \neg \neg \neg \phi \to \neg \neg \phi)&Axiom \ 1\\
3&\neg \neg \neg \neg \phi \to \neg \neg \phi&MP \ 1,2\\
4&(\neg \neg \neg \neg \phi \to \neg \neg \phi) \to (\neg \phi \to \neg \neg \neg \phi) & Axiom \ 3\\
5& \neg \phi \to \neg \neg \neg \phi & MP \ 3,4\\
6&(\neg \phi \to \neg \neg \neg \phi) \to (\neg \neg \phi \to \phi) & Axiom \ 3\\
7& \neg \neg \phi \to \phi & MP \ 5,6\\
8&\phi&MP \ 1,7\\
\end{array}
However, since the proof of $\phi \to \psi, \phi \to \neg \psi \vdash \neg \phi$ relies on Clavius, I took the road that I did.
Best Answer
If $T$ is an inconsistent set of first order theorems (or axioms for the proof), then it is possible to prove from $T$ for some $\alpha$ both $\alpha$ and $\neg \alpha$. So without the loss of generality we can assume that $T$ includes $\alpha$ and $\neg\alpha$.
Now suppose that $\beta$ is whatever first order sentence that you want to prove.
So you see, you can prove pretty much anything you want from $\\{\alpha ,\neg\alpha\\}$ for some first order sentence $\alpha$.