You need spatial data structures as part of computational geometry, to have at least some reasonable base reference against which to compare the time and memory consumption of special purpose (sweepline/randomized) algorithm. For this purpose, BVH is too vaguely defined to be useful (except for summarizing a common principle for some better specified data structures), but kD-trees are good enough for this purpose.
I don't see where you would need spatial data structures for non-geometry related data structures and algorithms, so I would say it is only a part of it, if you study geometry (or GIS) related problems. But if anybody shows me a non-geometry related application, then I'm quite willing to change my mind.
For geometric modeling, things like constructive solid geometry, homogeneous coordinates, Bézier curves, B-splines and NURBS should come first, because understanding them is also important when just using existing geometry modeling software. But BVH fits in well, because it is an intuitive geometric concept. It can sometimes be useful for understanding the performance behavior of existing geometry modeling software (sometimes the user can even influence the BVH to improve performance). I see fewer reasons why kD-trees should be part of geometric modeling.