lima, vc4: Don't force LINEAR modifier when PIPE_BIND_SHARED requested.
In case client app needs linear buffer, it will request (PIPE_BIND_SHARED | PIPE_BIND_LINEAR), otherwise using tiling/compression is more preferable for performance.
In case client app needs linear buffer, it will request (PIPE_BIND_SHARED | PIPE_BIND_LINEAR), otherwise using tiling/compression is more preferable for performance.