nvk: 2 buffer object alocation related optimizations
What does this MR do and why?
This Merge Request has two parts:
nvk: allocate cmd bos from the heap
[This leads to a significant performance improvement in some tests]
nvk/queue: Persistently keep the queue_state_bo
[Barely does much]
The rationale behind this is that by allocating less, and destroying buffers less will help performance.