Consider making `shape::Variant` an enum

As it currently is, there is some ambiguity for struct variants vs newtype variants containing a struct. This caused some issues with #2, but even with that fixed, I still feel like it should probably be handled separately. Also notable are unit variants vs. newtype variants containing units (although the difference there doesn't seem to be as much of an issue, since something like #2 could never happen with the unit variants).

Currently, although #2 is fixed, if you use an alias on a struct variant you'll end up with a name in the generated help text that doesn't match the alias you use. I would rather have it directly match the alias that is currently being used, instead of a random one being picked (which, due to how `serde_derive` handles aliases, often ends up being an alias instead of the original name).