Develop a parallel alternative for the slob allocator

[HanetakaChou]

Now the "gfx_connection_create_texture" consist of two stages. The first stage is serial due to the limit of the slob allocator. If we develop a parallel alternative, the streaming would be more efficient.