refactor: move wrap into separate attribute, cleanup derive macro code

[Frando]

This has a couple of refactorings on top of #51:

• Move the wrap functionality into a wrap attribute on variants, and adjust docs accordingly. I think this is better, it is a separate feature from the rpc types, and the rpc attr easily gets too long for a single line otherwise.
• Add support to set visibility and additional derives for wrap types. This makes this far more useful, otherwise you can't use the wrap argument as soon as you need a derive (e.g. Clone) on the generated type. By refactoring the code a bit, this barely adds more code to the macro.
• Make tx argument in rpc attribute optional, and change parsing to not use a map but just look for the only two arguments (rx and tx)
• Refactor the derive crate to be a bit simpler, and easier to follow by putting the main macro up to the top of the file
• Adds a full example to crate docs

The refactor of the derive crate has no changes apart from the wrap changes outlined above.

[rklaehn]

Can we just pass through all the things instead of specifically looking for derive(foo)? Maybe somebody wants to pass an attribute other than derive?

[Frando]

Let's do this once the need arises, ok? Not passing through everything reserves us the chance to add more of our own attributes.

[rklaehn]

I do like having wrap in a separate attribute.

Did we at some point have the option where the name of the struct is derived from the enum case? I seem to remember that, but maybe I was wrong. In any case, not sure if that is useful. Adding the name explicitly is probably less magic, which is good.