asahi: Add a BO cache (+ standalone fixes pulled from my other WIP branches)

The star of the show here is the BO cache, which is based on the BO cache in panfrost. Since BO allocation is slow, this improves performance considerably. The cache as it appears here is pretty well tested on Linux by now. Along with the BO cache, we have some memory usage improvements and general bug fixes. nir_op_pack_64_2x32_split support is a bug fix since it can be generated from the 64-bit lowering (hit this in an early version of UBO/VBO lowering).

Cc @lina