I don't know that one has ever been truly implemented, but it's certainly been discussed. Refer to just about any of Chris Date's work, primarily the third manifesto. I do recall reading about a database system that tries to implement the "true" relational model (mostly according to Date, with the exception of his type system), and has a language similar to Date's Tutorial D. I believe it was written in Haskell and showed up here not very long ago. But I can't find it now...
I don't know if Tutorial D and its brethren are actually better than SQL in the ways that SQL is bothersome to most people. I don't think most people actually care that SQL isn't a pure implementation of the relational system (and it goes further than Codd's original proposal, Codd called for NULLs, Date and Darwen want to throw them out). I think stuff like enabling code modularity and code reuse and otherwise acting like language design has advanced since the 1970s is far more interesting to most people than getting rid of bags in favor of sets.
For me when I tried to do something more complex and reusable tempating/generics were really missing, and lambda functions, but it didn't get into production postgres. Oh and better array support is needed..converting between arrays, select and values () is very tricky. So I was doing a lot of copy-pasting of my own tricks
There are people who are convinced that QUEL[1] is better (or at least was better at the time).
I never used it myself, so can't comment much about it. The examples on wiki makes it look as if perhaps it would integrate with code better, but that's just a guess.
Define "created". There's Datalog, LINQ, and Dedalus at the very least, but these aren't available for any widely used data store (except perhaps LINQ on some object database, like NDatabase).
