v3dv: refactor events

This makes it so we actually implement GPU-side event functions in the GPU using compute shaders, instead of having a purely CPU implementation that can cause submit stalls and incur in additional CPU-GPU sync.

Alternatively, we could have implemented events as syncobjects. Such implementation would require that we add a mechanism in the kernel to signal and reset syncobjects so we can queue these together with other GPU workloads and dependencies, but this would require changes to our current kernel API, which would mean that we would have to keep the old implementation around until we can assume a recent kernel version.