Related to issue #352, it doesn't fix things, but it's a start.

HumanBinaryData uses a more efficient "bytes" format when using non-human-readable serialisation formats (like CBOR). The deserialiser still uses deserialize_any – I haven't got a great way around this yet.

I've done a few things differently to Base64UrlSafeData; for example, this doesn't implement Display or FromStr – they're different to what Vec<u8> does, and we should move away from this (this will be the focus of #356).

It also implements Deref and DerefMut, which should make the type more transparent – and all of the common impls are in macros.

I've added bytes support to Base64UrlSafeData, so it can read what HumanBinaryData produces. Unfortunately this makes migration a two-step process.

While we're here, this makes serde_json a dev-dependency of base64urlsafedata, because it's only used in tests; and this also adds a bunch more tests for expected input formats.

• cargo test has been run and passes
• documentation has been updated with relevant examples (if relevant)