-HDR isn't a pass. Tonemapping is the part of HDR that's implemented as a postprocess.
-you're more or less "correct" already, as you should do most effects in HDR. You'll pay for it though.
-to fold together passes, just look at it in general (we can't say exactly HOW you'll implement things). Probably vignetting/gammacorrecting can be stuffed into another pass. But since all your convolutions depend on each other, you'll have to store the results somewhere.
Well, if you've got your level, and k=2, you're there
since an object fits inside the node, any bounding box with it's centre in the node will fit, if you calculated the level correctly. (worst case scenario: centre of bb on the edge of "tight" node, bb's dimensions is equal node's dimensions. Half of the bb is outside the "tight node", but each side of the "tight" is padded by the same amount (half dimension) of loose.
or, in pseudo
int depth; //you got this
int horizontalNodeIndex = depth*(bb.centre.x / sceneBB.width)