Mr. McIlroy, FWIW I would love to read more about that McCarthy talk. It sounds like an amazing experience.
There is no authoritative source about these pragmas. They are listed and described in the GHC User Guide, but that source all too often defines solely by example, not even bolstered by a formal syntax specification.
I think it would very helpful simply to better (and more rigorously) document the syntax and semantics of the available extensions. There is currently a call to action to update GHC's Haddock documentation in preparation for the 8.0 release [1]. Perhaps some effort can also be directed towards the documentation of the LANGUAGE pragmas. There is still a problem, though: For completeness, one must consider the interactions of the various subsets of these pragmas, some of which are already known to be unsound. What's worse, the number of extant pragmas already makes an enumeration of these subsets impractical, since there are some 10^31 of them even ignoring the "NoX" pragmas. The only long-term solution then seems to be to codify a new Haskell standard that incorporates some known-good subset of these pragmas that the community seems to agree on, which I suppose is part of the task that the Haskell Prime committee has before them. I do not envy them. [1]: https://mail.haskell.org/pipermail/ghc-devs/2015-December/010681.html