* muxes N AXI-MM masters to one AXI-MM slave
* read and write channels are mux'ed independently
* no interruptions during bursts, next schedule on LAST
* address valid is used to signal transfer requests
* may cost up to N-1 cycles latency