a) Regular implies normal.
Okay, maybe that's too high-tech. If $f=g+hy$ lies in the integral closure, then it satisfies the quadratic polynomial $(z-g)^2 - h^2(x^3-x)$. If the coefficients are polynomials (which they must be, by Gauss's Lemma), then $g$ is a polynomial, and $h^2 (x^3-x)$ is a polynomial. But $x^3-x$ is square-free (here we use that the characteristic is not $2$), so $h$ is a polynomial and $f\in R$.
b) Since $y^2,x^3 \in P^2$, it is straightforward that $x\in P^2$ and $P^2 = (x)$.
c) (Note: it is sufficient to address this problem for $R\otimes_k \overline{k}$, so we assume that $k$ is algebraically closed to avoid some technicalities.)
First of all, $X=\operatorname{Spec}k[x,y]/(y^2-x^3+x)$ is an open subset of an elliptic curve $\overline{X} = Y\cup \{\infty\}$. This is a rather important observation, because it implies that $R=k[x,y]/(y^2-x^3+x)$ doesn't have any principal prime ideals whatsoever (except, of course, for $(0)$).
We will demonstate this, real mathematician-style, by looking at the global behavior of a hypothetical generator of $P$. If $P=(f)$ is principal, then $f$ extends to a rational function on $\overline{X}$ with a simple pole at $\infty$. Therefore $1/f$ is a non-constant rational function on $\overline{X}$ with only a simple pole at $P$. This implies that $l(P) \geq 2$.
Here $l(P)$ is shorthand for the dimension of the space of rational functions on $\overline{X}$ which have at worst a simple pole at $P$, and are regular elsewhere. In general, $l(D)$ is defined for $D$ a Weil divisor, i.e. a formal sum of points. For example, $l(P+2Q-3R)$ counts the number of linearly independent functions which must have a triple zero at $R$, and are allowed to have a pole of order $1$ at $P$, and $2$ at $Q$. The sum of the coefficients of $D$ is called its degree.
We will calculate $l(P)$ exactly using the Riemann-Roch theorem, which says that, for any Weil divisor $D$, $l(D)-l(K-D) = \operatorname{deg}{D} + 1 - g$, where $g$ is the genus of $\overline{X}$, and $K$ is the divisor associated to any differential form on $\overline{X}$. For an elliptic curve, we must have $K=0$ and $g=1$, so this equation takes the simple form $l(D)-l(-D) = \operatorname{deg}{D}$.
We set $D=P$. Since $\operatorname{deg}{P} = 1$, and $l(-P)=0$ (in fact, $l(-D)=0$ for any divisor of positive degree), we conclude that $l(P) = 1$. Since earlier we found that $l(P)\geq 2$ for any principal maximal ideal on $X$ (in fact, the proof works for any projective curve with a single point removed), we have our contradiction.
(The above does have a nice, concrete interpretation: when you strip away the machinery, the argument is that, if $P=(f)$, then the map $k[x]\to R$ sending $x$ to $f$ is an isomorphism. We use basic facts about divisors to avoid some difficult calculations, and genus to show that $R$ cannot be a polynomial ring.)