The "complete the square" method allows you to center the conic, i.e. make the linear terms vanish (by a translation, $ax^2+bx+c\to a'u^2+c'$).
To deal with "obliqueness", you need to let the cross term $xy$ vanish, by means of a rotation.
Let $x=cu-sv,y=su+cv$, which expresses a rotation around the origin ($c,s$ denote the cosine and sine of the angle), and let the centered equation be in the form
$$Ax^2+2Bxy+Cy^2=1.$$
Then substituting,
$$(Ac^2+2Bcs+Cs^2)u^2+(-2Acs+B(c^2-s^2)+2Ccs)uv+(As^2-2Bcs+Cc^2)v^2=1.$$
By a suitable choice of the angle, you can achieve
$$-2Acs+B(c^2-s^2)+2Ccs=0$$
and the equation reduces to
$$A'u^2+C'v^2=1.$$
Depending on the signs of $A',C'$, you get an ellipse or an hyperbola. By a further rescaling of the variables, you can obtain the "canonical" forms (circle and equilateral hyperbola)
$$p^2\pm q^2=1.$$
To find the suitable angle, use the "double angle" formulas and rewrite
$$(C-A)\sin(2\theta)+B\cos(2\theta)=0,$$ which is easy to solve.
There are two ways to prove this:
Formal:
You can show, through a bunch of ugly computation, that the expression $B^2-4AC$ is invariant under rotation. So, consider when $B=0$ (in other words, when the conic section's directrix is parallel to one of the axes). It is easy to see that for a hyperbola $-4AC$ is positive, for an ellipse $-4AC$ is negative, and for a parabola $-4AC$ is $0$. For a better worded explanation, go to this link.
Very Informal But Intuitive:
Take the equation $Ax^2+Bxy+Cy^2+Dx+Ey+F=0$. Imagine if $x$ and $y$ were very large numbers. We can forget about $Dx+Ey+F$ because it becomes insignificant when compared to $Ax^2+Bxy+Cy^2=0$. Now, divide by $x^2$:
$$A+B\left(\frac{y}{x}\right)+C\left(\frac{y}{x}\right)^2=0$$
We notice that we now have a quadratic in $\frac{y}{x}$.
This next part is a little hard to explain in words (and my English is sort of bad) but I will try my best.
The number of solutions to this equation represents the number of ways in which the graph of the equation "zooms off towards infinity." Imagine zooming out really far from a graph of a hyperbola. You would only see an "X" formed by two lines (these lines are the asymptotes of the hyperbola). If you solve for $\frac{y}{x}$ in the above equation, you would be solving for the slopes of those lines. Imagine zooming out really far from a graph of a parabola. You would only see one line (the axis of symmetry for the parabola). If you solve for $\frac{y}{x}$ in the above equation, you would be solving for the slope of that line. If you zoomed out really far from a graph of an ellipse, you would see a point.
So, if $A+B\left(\frac{y}{x}\right)+C\left(\frac{y}{x}\right)^2=0$ has two solutions for $\frac{y}{x}$, the equation is a hyperbola. One solution means parabola. Zero solutions means ellipse or circle. The number of solutions corresponds to the sign of $B^2-4AC$.
I sort of like this informal proof because it explains why the discriminant of a conic looks like that of a quadratic.
Best Answer
Such a curve having the shape of a double-sided axe (https://en.wikipedia.org/wiki/Labrys) with mixed elliptical/hyperbolic boundaries can be given by the general equation
$$x^2+k|y^2-1|=k^2 \tag{1}$$
with a "behavorial change" when variable $y$ crosses values $y=1$ or $y=-1$ as can be seen on the following graphics for different values of $k$:
Fig. 1: Curves with equation (1) for different values of $k$, with $k=0.25$ for the most internal one (in two parts) and $k=2$ for the most external, with steps $0.25$. For our question, only values $k>1$ (blue curves) make sense.
It remains now to tilt the "axe" using the following coordinates change in (1):
$$\begin{cases}x&=& \ \ \ (\cos \theta) X + (\sin \theta) Y \\ y&=&-(\sin \theta) X + (\cos \theta) Y \end{cases}$$
(It looks like a pun: a change of axes inducing a different axe...)
Explanations:
I have obtained (1) by imposing common foci to the ellipse and the hyperbola. How can it be done ? If $f$ is the common distance from the origin to the foci, it is known that $f^2=a^2-b^2$ for an ellipse and $f^2=a^2+b^2$ for a hyperbola with canonical equations $\dfrac{x^2}{a^2}\pm\dfrac{y^2}{b^2}=1$ ; one obtains in particular $f=\sqrt{k^2-1}$.
As a consequence, the elliptical arcs are orthogonal to the hyperbolic arcs. This wasn't necessary (see the solution by @David K).
Case $k=1$ is particular as can be seen on Fig. 1 : the elliptical arcs become circular arcs.
Edit: Here is the Matlab program I have written for the generation of the figure: