First, the inversion of the outputs simply means that the output is active low. That is, for an input of 0000, the 0 output is selected, and it is driven low. All the other ouputs stay high. The NAND gates are used because, given that the active lines on the 74154 are low, DeMorgan's Theorem allows NAND gates to function as OR gates. That is, if the 74154 outputs were active high, OR gates would perform the synthesis desired. Since the ouputs are active low, NAND gates do the job.
The active-low enable inputs allow cascading of demultiplexers over many bits. If you wanted to generate a 1 of 256 demultiplexer, you could use 16 74154s looking at the 4 least significant bits, while a single 74154 would look at the 4 most significant bits, with one ouput going to each of the other 16 74154s. And why are there 2 of them, you ask? It's because with a 24-pin package, 2 power pins + 4 address pins + 16 output pins = 22. Rather than providing only a single enable, both pins are used. This allows more flexibility in the logic functions available.