the only thing i can think of seems really inefficient
some advanced rendering engine probably has it figured out using advanced techniques
since i dont know anything about this stuff i would do something really simple like making a table that contains every object and all the points of the object and figure out the 'space occupied' in xyz space
then do a calculation to make sure all the objects' boundaries don't intersect, if they do move them until they dont, something like that

precomputed binary space partition if they're static
viewport depth buffer if dynamic

>>110246294
Assuming top-down ensure every pixel for the depth buffer write equals the sprite's Y position (provided as an attribute) instead of the quad's actual depth
>>110246344
>>110246347
wtf lol