Deduplicate new puzzles by URL.

[jpd236]

When adding a puzzle, we start a transaction to atomically check if a puzzle already exists with that URL and only add it if either no such puzzle exists or if the client indicates that duplicate URLs are allowed. In the event of a transaction failure, we delete the Google document that was created optimistically before saving the puzzle.

On the client side, we add a new checkbox to allow duplicate URLs, unchecked by default. If we encounter a duplicate URL error, we show an error indicating that the puzzle may be a dupe and telling the user what to check if the duplicate URLs are intentional.

See #2275

[jpd236]

This should be functional, but there are some less-polished (IMO) elements as I wanted to get feedback before moving forward. At least the following:

• The UX. Do we always want to show this checkbox as is done here? Or should it be a more interactive flow, e.g. a (nested) dialog indicating that the puzzle is potential duplicate and offering the user a simple Continue/Abort option? Should we provide more details about the duplicate? I kept it simple for now.
• The error passing. I just throw a Meteor.Error with an unusual (but reasonable) HTTP error code and detect it client side. This should probably be something a bit more semantically meaningful. What's the canonical way of doing this?
• The transaction. This seems like the first use of a transaction in the codebase. It seems reasonable at the face of it, but I'm pretty unfamiliar with MongoDB and am not sure about the semantics. In other places, we just use locks - that could also be reasonable here, e.g. with a hunt-specific puzzle creation (and updating?) lock, if that's preferable.