It depends how you want to visualize the data. Tools like imshow() or imagesc() represent the data points at the center of each facet in the displayed image. Tools like pcolor() represent the data at the vertices. If the latter is what you intend, then just use pcolor() instead of imagesc().
If you actually want the matrix represented as an image, then you'll have to create the grid. One way would be to just overlay a mesh plot.
dx = diff(xrange)/(s(2)-1);
dy = diff(yrange)/(s(1)-1);
xg = linspace(xrange(1)-dx/2,xrange(2)+dx/2,s(2)+1);
yg = linspace(yrange(1)-dy/2,yrange(2)+dy/2,s(1)+1);
hi = imagesc(xrange,yrange,a); hold on
hm = mesh(xg,yg,zeros(s+1));
You might try to do it by setting the grid properties of the axes, but you'll have to contend with the ticklabels if you do that.
As to how you make the added grid appear only at certain zoom levels, you'd have to create some sort of custom callback to do that. I'll leave that to someone else.