Draft: nvk: Implement VK_EXT_mesh_shader
This implement VK_EXT_mesh_shader on NVK.
This is based on the mixin patches from !25668 (closed).
The following things needs more work:
- Emulation for more than 32 local invocations needs to be rewritten to respect barriers.
- shared_atomic for xchg/cmpxchg/fcmpxchg needs to be lowered properly. (and probably same story for task_shared)
- Task shaders need more testing. (there seem to be some issues with the global invocation index for example but I haven't looked much more into that)
Closes #9641
Edited by Mary Guillemard