Normal Vector to a Sphere

This is a typical example of singular point(s) arising from the parametrization, they are called the artificial singularities.

For a parametrized surface: $S: \mathbf{r} = \mathbf{r}(u,v)$, like you said: $$ \mathrm{If } \;\; \mathbf{r}_u \times \mathbf{r}_v \neq 0, $$ then it indeed is a normal vector (un-normalized) to the surface $S$. Now: $$ \frac {\partial P(\phi, \theta)}{\partial \phi} \times \frac {\partial P(\phi, \theta)}{\partial \theta} = \begin{vmatrix} \mathbf{i}& \mathbf{j} &\mathbf{k} \\ \cos\phi \cos\theta & \cos\phi \sin\theta & -\sin\phi \\ -\sin\phi \sin\theta & \sin\phi \cos\theta & 0 \end{vmatrix}, $$ notice the cross product is 0 for $\phi = 0,\pi$, this happens due to the choice of parametrization. For example, if you choose the angle between $y$-axis as the polar angle, then this singular point will be gone (other arises though), and this gives you a heuristics of one of the reasons why we wanna cut a manifold into pieces and establish a local coordinate system...

For a sphere, the surface normal is exactly your $P(\phi,\theta)$, since the normal is just the vector from the origin. Your vector $\vec N$ is indeed normal to the surface, but it's not normalized: $$|\vec N| = |\sin \phi \cdot \vec P| \neq1$$ The correct definition of the surface normal , coupled with the limit $\theta \to 0$ should give you the correct result: $$\vec N=\frac{\vec a \times \vec b}{\sqrt{|\vec a|^2|\vec b|^2-|\vec a\cdot\vec b|^2}}$$ Where: $$\vec a = \vec P_\phi, \ \ \vec b =\vec P_\theta$$