brw/anv: new push constant mechanism for compute/task/mesh stages
What does this MR do and why?
For device generated commands we need a more flexible push constant mechanism : load push constants through A64 messages
Take the opportunity to update mesh/task shaders to use the same loading mechanism as compute on Gfx12.5+.
This MR currently hangs on a DG2 CI system (on Enable ReBAR fixed it. Unfortunately it was a hard hang, not recovering the GPU.brw/anv: rework push constants for mesh/task shaders
, MTL/LNL are fine), but it could be an issue related to the system not having ReBAR support.