`String` successfully parses bytes buffer

I have a pre-existing msgpack schema that I tried to encode in rmp-serde like this:

```
#[derive(Serialize, Deserialize, PartialEq, Debug, Clone)]
#[serde(untagged)]
pub enum Value {
Numeric(f32),
String(String),
#[serde(with = "serde_bytes")]
Bytes(Vec),
Bool(bool),
}
```

However, if I use `rmp-serde` to decode a buffer that happens to contain valid utf-8, like

```
{
"m": "set",
"path": "ui-state",
"value": {
"type": "Buffer",
"data": [
97
]
}
}
```

(base64 `g6Fto3NldKRwYXRoqHVpLXN0YXRlpXZhbHVlxAFh`), I get a `String`, but I actually wanted a `Bytes`. It would be nice if there were some sort of strictness annotation that would make this behave as intended.