Test for internal strictness

[treeowl]

#232 revealed that Data.HashMap.Strict.insert could produce a HashMap containing "stupid thunks" (i.e., ones guaranteed to compute successfully in O(1) time). @mantasg discovered this bug in the process of playing around with ghc-heap-view. Could we use ghc-heap-view (or borrow its techniques) to check for these problems? I believe we want to maintain the invariant that for any function in Data.HashMap.Strict, if none of the HashMap arguments to the function contains a thunk, then the result will not contain a thunk. In particular, a HashMap built solely using functions in Data.HashMap.Strict should never contain a thunk.

[sjakobi]

Maybe we can use nothunks for this.

I have recently discovered some space leaks and broken strictness properties (#379, #381, #382, #383) just by reading the code and generated Core. There may be more subtle issues lurking.