Which reasons were discussed for potentially changing the proof-of-work algorithm to something else in the future (Cuckoo cycle in particular)?
One of the main reasons Cuckoo cycle is atractive when compared to Cryptonight is that it is very fast to verify. This is one drawback of Cryptonight: it makes all operations that need to verify Cryptonight hashes slower.
Another reason to want a switch to Cuckoo cycle would be to keep the CPU/GPU/ASIC performance within a reasonable scale. Should ASICs pop up with enough memory allowing them to leave CPUs and/or GPUs behind by too high a margin, this would be undesirable from a decentralization point of view.
A more obscure reason to prefer Cuckoo cycle is that it is difficult to make a similar pool mining setup to bitcoin or monero. What these do is get miners to find a share at a lower difficulty than needed for a block, because a valid block solution will also be a valid share. For Cuckoo cycle, this is not the case, so a miner can't periodically send a "proof of work" to the pool. It is not quite clear whether this particularity is wanted or not. While it will force miners to solo mine (increasing decentralization), many might decide to stop mining instead (decreasing overall network security).