Personally, I'd think Hackage getting some kind of support for popularity metrics would go some way to solving this problem, by helping create a consensus of what packages used the most by the Haskell community.
http://www.reddit.com/r/haskell/comments/o5iov/modern_reallw...
@peaker I think each of your issues has prior art. The author of yesod has had his gotcha/critical-list, and a response from a Simon!
http://www.reddit.com/r/haskell/comments/k4lc4/yesod_the_lim...
and partial functions in prelude
http://www.reddit.com/r/haskell/comments/lf71l/deprecate_pre...
http://haskell.1045720.n5.nabble.com/Proposal-Move-the-insta...
re: redoing Num and numeric tower (prolly won't happen)
http://www.reddit.com/r/haskell/comments/nt0op/i_wrote_a_mod...
also: quality/popularity of packages, as more and more hit git/github, they could post #s of forks, pulls etc, as a crude proxy for ... something
* Marking partial functions with some naming convention
* Polymorphic containers, with polymorphic "length" and other features people take for granted
* Removing monomorphic functions such as "map" and "empty" -- expose APIs almost always as part of generic classes
* type String = Text (Along with extra polymorphism it should not cause any trouble)
* Sane numeric hierarchy that matches the mathematical definition
* Fix the Monad/Applicative duplication mess (Also remove "fail" from Monad)