Adds Mesh equivalents of `multimesh_get_buffer_rd_rid` so GPU-driven mesh
workflows in a GDExtension can write directly into a surface's vertex,
attribute, skin, or index buffer from a compute pipeline, avoiding the
CPU round-trip through `mesh_surface_update_*_region`.
A new `ARRAY_FLAG_USE_STORAGE_BUFFER` surface format flag opts the backing
buffers into storage-buffer usage so they can be bound in compute shaders.
Vulkan/D3D12/Metal only, OpenGL returns an invalid RID.
Kotob M
7cbebab1d0