Simplify enum representation in serde_test

[dtolnay]

Continuing the discussion from #8:

I think it'd be best to serialise enums with multiple tokens so that Enum is used always, rather than making variants include names and not have the start header. I think it'd be cleaner and make it easier to understand the purpose of the token rather than duplicating the information over multiple token types.

@clarcharr

[dtolnay]

Here are the reasons I hesitated to eliminate the Token::*Variant form:

1. The more flexible form is only rarely needed. In our test suite right now we have 7 uses of Token::Enum to 54 uses of Token::*Variant. And of the 7, really 4 of them could be written using Token::NewtypeVariant so the balance is 3 to 58.

   $ git rev-parse HEAD
     5415abf80bb3214498f484ed08d7bace561aa5fc
   $ rg -c 'Token::Enum\(' test_suite/tests/
     test_suite/tests/test_de.rs:7
   $ rg -c 'Token::(Unit|Newtype|Tuple|Struct)Variant\(' test_suite/tests/
     test_suite/tests/test_de.rs:8
     test_suite/tests/test_ser.rs:6
     test_suite/tests/test_macros.rs:22
     test_suite/tests/test_annotations.rs:18
   

2. I think Token::*Variant is a better match for people used to the Serializer API which is broken down into serialize_unit_variant / serialize_newtype_variant / serialize_tuple_variant / serialize_struct_variant methods.

3. Token::Enum as it exists right now is ambiguous, which makes it really only useful for testing deserialization and not for serialization.

   [
     Token::Enum("E"),
     Token::Str("A"),
     Token::Unit,
   ]

   Is this a unit variant E::A or a newtype variant E::A(())?