Refactor Model/Handler distinction

[pbrisbin]

This PR has two motivations:

First, see if it is as easy/safe to wildly refactor a Haskell codebase as we keep saying it should be. It was easy, we'll have to see if it was safe.

Second, bring the codebase in line with a few best practices, some accepted (line-length, formatting, import sorting, etc) and some that I'm proposing now via this PR.

The new ones are:

• Separate Model.* modules for each non-trivial data type
  • JSON instances (even though they're orphans, I think it's the best place for them)
  • "virtual accessors" i.e. userName
  • Query functions, which must return DB a and cannot mention Handler
• Anything dealing with Request/Response should appear in Handler.* or a Helper.*
• Helpers should be about cross-cutting concerns, and not be model or handler-specific
• One Handler module per resource, not per URL

[jferris]

I think I commented on some of this on the other pull request, but this model approach looks good to me.