This is not quite a solution of the original problem, but rather of a related and, perhaps, even more natural one: 

> How small can the ranks of real matrices $B=(b_{ij})_{1\le i,j\le n}$ and  $C=(c_{ij})_{1\le i,j\le n}$ be, given that $b_{ii}c_{ii}\ne 0$, while $b_{ij}c_{ji}=0$ whenever $i\ne j$?

The inequality $\rk(B\circ C)\le\rk(B)\rk(C)$ gives 
> $$\rk(B)\rk(C)\ge n,$$ 

and I claim that this estimate is sharp. To see this, associate the rows and columns of our matrices with the elements of a finite abelian group $G$ of order $|G|=n$, consider a direct sum decomposition $G=G_1\oplus G_2$ and, denoting by $I_1$ and $I_2$ the indicator functions $G_1$ and $G_2$, respectively, for $u,v\in G$ let $b_{uv}:=I_1(u-v)$ and $c_{uv}:=I_2(u-v)$. We have then $b_{uv}c_{vu}\ne 0$ if and only if $u-v\in G_1$ and $v-u\in G_2$, which is equivalent to $u=v$. On the other hand, each row of $B$ is then identical to $|G_1|$ other rows, showing that $\rk(B)\le|G|/|G_1|=|G_2|$ and, similarly, $\rk(C)\le|G_1|$.

Thus, for instance, for $n=9$, taking $G=\mathbb Z_3^2$, the matrices may look as follows:
  $$ B=\begin{pmatrix}
         1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\ 
         1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\ 
         1 & 1 & 1 & 0 & 0 & 0 & 0 & 0 & 0 \\ 
         0 & 0 & 0 & 1 & 1 & 1 & 0 & 0 & 0 \\ 
         0 & 0 & 0 & 1 & 1 & 1 & 0 & 0 & 0 \\ 
         0 & 0 & 0 & 1 & 1 & 1 & 0 & 0 & 0 \\ 
         0 & 0 & 0 & 0 & 0 & 0 & 1 & 1 & 1 \\ 
         0 & 0 & 0 & 0 & 0 & 0 & 1 & 1 & 1 \\ 
         0 & 0 & 0 & 0 & 0 & 0 & 1 & 1 & 1
       \end{pmatrix}, \quad
     C=\begin{pmatrix}
         1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 \\
         0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 \\ 
         0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 \\ 
         1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 \\
         0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 \\ 
         0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 \\ 
         1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 \\
         0 & 1 & 0 & 0 & 1 & 0 & 0 & 1 & 0 \\ 
         0 & 0 & 1 & 0 & 0 & 1 & 0 & 0 & 1  
        \end{pmatrix}. $$  

I still wonder whether it is possible to modify somehow the construction to have $B=C$ and $\rk(B)=\rk(C)\approx\sqrt n$.