Currently for FILL_SPANS opcode, all the spans are put into the RenderBuffer and QueueFlusher takes the spans from RQ and call MTLRenderer. If the spans numbers are large, the FILL_SPANS opcode will be sent several times.

We can optimise by storing the spans in a separate buffer and add pointer to buffer into the RQ so that we can get 1 FILL_SPANS with pointer to span data.