First of all a short remark: I cannot really answer any of the three questions you asked at the end of your post. (For example, I have also the impression that it is not immediately clear why $P+N,PN \in K$.)
Nevertheless, here is my reconstruction of the proof you asked for. In some details, it is different from the proof you outlined. Instead, my proof will be in accordance with the answer to your question on mathoverflow (In fact, my proof hopefully answers the questions you asked in the comments to the mathoverflow-answer.)
Similar as in the proof you outlined, we can also write
$$\Delta_K=(\sum_{\pi \in X} \prod_{i=1}^n \sigma_{\pi(i)}a_i-\sum_{\pi \in Y} \prod_{i=1}^n \sigma_{\pi(i)}a_i)^2=(P-N)^2$$
where $X$ is the set of even permutations and $Y$ is the set of odd permuations in $S_n$. Note that we can assume $K \subset \mathbb{C}$. Let $L$ be a field with $K \subseteq L \subseteq \mathbb{C}$ and such that $L/ \mathbb{Q}$ is finite and Galois. (It is a well known fact in Galois Theory that a suitable $L$ exists). I will only explain the "critical step" of the proof, namely that $P+N, PN \in \mathbb{Q}$. (The rest should then be clear.) First of all I show the following statement:
($\star$)$\phantom{aaaaaaa}$$\phi(P+N)=P+N$ and $\phi(PN)=PN$ for all $\phi \in \text{Gal}(L/\mathbb{Q})$
Let $\phi \in \text{Gal}(L/\mathbb{Q})$. It follows from the normality of $L$ that $\sigma_iK \subseteq L$ for all the embeddings $\sigma_i$. Therefore, all the $\sigma_i a_j$ are in $L$ and we can apply $\phi$ to those elements (and thereby to $P$ and $N$). Moreover, for every $\sigma_i$, we can build the composition $\phi \circ \sigma_i$ and this is an embedding $K \to \mathbb{C}$. Now, it is not hard to see that the association $\sigma_i \mapsto \phi \circ \sigma_i$ gives us a bijection
$$\{\sigma_1,...,\sigma_n\} \to \{\sigma_1,...,\sigma_n\}$$
But this means that we can find a permutation $\tau \in S_n$ such that for every $i \in \{1,...,n\}$
$$\phi\circ \sigma_i=\sigma_{\tau(i)}$$
Distinguish two cases concerning $\tau$.
Case 1: $\tau$ is even. Then, $\tau X=X$ and we have
$$\begin{align}
\phi(\sum_{\pi \in X} \prod_{i=1}^n \sigma_{\pi(i)}a_i)
& = \sum_{\pi \in X} \prod_{i=1}^n \phi\circ\sigma_{\pi(i)}a_i \\
& = \sum_{\pi \in X} \prod_{i=1}^n \sigma_{\tau\pi(i)}a_i \\
& = \sum_{\pi \in \tau X} \prod_{i=1}^n \sigma_{\pi(i)}a_i \\
& =\sum_{\pi \in X} \prod_{i=1}^n \sigma_{\pi(i)}a_i
\end{align}$$
This shows $\phi(P)=P$. In a similar fashion, you can show that $\phi(N)=N$ (the key point is here that $Y=\tau Y$.)
Case 2: $\tau$ is odd. Here we have $\tau X=Y$ and $\tau Y=X$. It follows that $\phi(P)=N$ and $\phi(N)=P$ (the proof is similar to the one in case 1: simply write down the formulas for $\phi(P)$ and $\phi(N)$ and during the calculations, make use of $\tau X=Y$ and $\tau Y=X$).
In every case, $\phi(P+N)=P+N$ and $\phi(PN)=PN$ and this shows ($\star$).
Now, $(\star)$ says exactly that $P+N$ and $PN$ are in the fixed field of the Galois group. But this means $P+N, PN \in \mathbb{Q}$.
Best Answer
A homogeneous system of $n$ linear equations wth $n$ unknowns has a nontrivial solution iff the determinant of the system is $0$. That is the statement used in the OP.