> I never understood there is a relationship between quadratic equations and some kind of underlying mathematic geometric symmetry.

In a polynomial equation, the coefficients can be written as symmetric functions of the roots: https://en.wikipedia.org/wiki/Vieta%27s_formulas - symmetric means it doesn't matter how you label the roots, because it would not make sense if you could say "r1 is 3, r2 is 7" and get a different set of coefficients compared to "r1 is 7, r2 is 3".

Since the coefficients are symmetric functions of the roots, that means that you can't write the roots as a function of the coefficients - there's no way to break that symmetry. This is where root extraction comes in - it's not a function. A function has to return 1 answer for a given input, but root extraction gives you N answers for the nth root of a given input. So that's how we're able to "choose" roots - consider the expression (r1 - r2) for a quadratic equation. That's not symmetric (the answer depends on which one we label as r1 and which we label as r2), so we can't write that expression as a function of the coefficients. But what about (r1 - r2)^2? That expression IS symmetric - you get the same answer regardless of how you label the roots. If we expand that out we get r1^2 - 2r1r2 + r2^2, which is symmetric, which means we can write it as a function of the coefficients. So we've come up with an expression whose square root depends on the way we've labeled the roots (using Vieta's formulas you can show it's b^2-4c, which you might recognize from the quadratic equation).

Galois theory is used to show that root extraction can only break certain types of symmetries, and that fifth degree polynomials can exhibit root symmetries that are not breakable by radicals.