For some optimizations coming on the userspace side, splitting larger
draw or gmem cmds into multiple cmdstream buffers, we need to support
much more than the previous small/arbitrary limit.
Signed-off-by: Rob Clark <robdclark@gmail.com>