If this were a core feature that many developers were anxiously awaiting, I could understand the desire to release and iterate. But is there really a pressing need for an IP library in the stdlib?
You should have asked this question a few month ago. Now, release mechanics make it difficult to remove the library, except if a severe problem was uncovered - which you haven't been able to demonstrate.
I don't believe that your issue "hosts and nets are represented with the same class" is severe: it is very well possible to use the IP function to represent individual hosts (including having a string representation that doesn't print a netmask). The only possibly-missing feature is support for rejecting host strings that include a mask on parsing; that can be added in a backwards-compatible way as an optional parameter to the IP function (as discussed on the tracker).