EDIT: Keeping it here because I don't like deleting stuff, but you really should disregard this as it A) fails to notice the 1-point compactification (I don't know what came over me...), and B) doesn't answer the problem of keeping a manifold structure.

This is a completely elementary approach that seems to work:

If $(x,x)$ is a point in the diagonal and $(x_n)$ and $(x_n')$ are "essentially different" sequences(*) that converge to $x$, then clearly any compactification of $S_2 \times S_2 \setminus \Delta$ has to contain $(x,x)$. And since $S_2 \times S_2$ is _a_ compactification of $S_2 \times S_2 \setminus \Delta$ and it contains all the points it absolutely has to to be a compactification, it must be the smallest one.

As to whether there are larger ones... probably, but it's fairly obvious that $S_2 \times S_2$ is the Stone–Čech compactification of $S_2 \times S_2 \setminus \Delta$, so that puts some pretty clear limits on what other interesting compactifications you can have.

(*): More formally we could write $x(\mathbb{N}) \cap x'(\mathbb{N}) = \emptyset$.