Rethink haystack problems and innovate on shortcomings

There are a few open questions with haystack problems:

- How do we ask users to find important information in messy descriptions and expect _one correct_ implementation? (note this means we have to write non-descriptive functions names -- modeling bad practice to users).

Could we not ask them to implement the types signature of the function? And write input and output tests? I guess we really have no guarantee about what they write... Maybe this does just need human evaluation