Transcoders applied to multi-layer perceptron layers predominantly recapitulate individual neurons or sparse combinations of co-activating neurons