As of today, (November 2018):
To what extent are those criticisms of Lightning valid as of November 2017?
This is still a case for Lightning. Eltoo was the first to suggest a simpler way of coping with such storage requirement. They do not require storing keys or older transactions. Their disadvantage is that there is no punishment in the event of fraud, leading to selfish users always publishing the transaction that maximises their balance. In current Lightning Implementations, it is important to keep track of Breach Remedy Transactions, and of course never lose new Commitment Transactions (as this can incur in a punishment to oneself after publishing an old balance unknowingly).
Is there an implementation of Duplex micropayment channels or at least plans to create one?
Current implementations are focusing in Lightning Channels for two main reasons:
1. Eltoo channels are too recent (perhaps somebody is working publicly in an implementation, I am not aware)
2. DMC channels have a trade-off between the utility of the channel and the worst-case collateral cost of the channel, that discourages its use. This is even more important when considering n parties (such as factories). For more information about this trade-off, see the results in this document
DMCs are however better when one knows of a period of time that one will be offline, so I believe it will be implemented in the future. But there is no current implementation in the main ligthning development teams.