B → K* MX vs B → KMX as a probe of a scalar-mediator dark matter scenario