[Nouveau] Using one buffer object per (EXA) pixmap potentially wastes memory.

[Maarten Maathuis]

I've been doing some testing and it seems that glyphs (typically
smaller than a page size on nv50) render slow'ish sometimes, mostly
due to the many trips to the kernel (drm_addmap_core is the symbol
that shows up). Ofcource this was a benchmark, but it does leave me
wondering, do we really need need to call the kernel for every little
pixmap? Also from a memory point of view, because a single page can
theoretically contain 32 pixmaps. Is it problematic to combine small
allocations into at least a single page?

Maarten.