I'm not going to respond to the part about dealing with prepared statements errors, since I think we've already covered that and there's nothing new being said. I don't find automatic savepointing acceptable, and a significant change of the PostgreSQL protocol to support this doesn't seem reasonable (but you can try proposing).
As mentioned before. JDBC is not the only postgres driver to do this the ODBC driver does this as well. This is a requested feature by users. We didn't just decide to do it on our own.
One thing to keep in mind is that both JDBC and ODBC are not exclusively PostgreSQL drivers and as such we sometimes have to jump through hoops to provide the semantics requested by the API.
I don't have anything in particular against automatic savepointing when requested directly by users (especially if it's opt-in). My problem is more with automatic savepointing as a solution to a problem created by doing automatic prepared statements... Way too much automatic stuff going on there for my taste...