Fix: Split serde bound for RaftError into serialize and deserialize

[tvsfx]

While implementing a struct that has a generic RaftError<E> as a field, I ran into some issues with trait bounds because RaftError does not make a distinction between the serde Serialize and Deserialize bounds. The consequence is that any impl block that requires my struct to be Serialize will require E to both be Serialize and DeserializeOwned, which is not necessary. This PR splits the two bounds.

Checklist

• Updated guide with pertinent info (may not always apply).
• Squash down commits to one or two logical commits which clearly describe the work you've done.
• Unittest is a friend:)

---

This change is 

[drmingdrmer]

Oh thank you! I've not considered such subtle scenario yet.

[tvsfx]

FWIW, I was confused by the presence of this bound, so I looked into the way this concrete derivation works. I think the problem is not the bound on the E generic, but rather, the fact that NID is receiving two different Serialize/Deserialize bounds; one from the derive annotation, and one from the NID: NodeId trait bound. The following hence seems like a cleaner fix to me:

#[derive(Debug, Clone, PartialEq, Eq, thiserror::Error)]
+#[cfg_attr(feature = "serde", derive(serde::Deserialize, serde::Serialize))]
-#[cfg_attr(
-    feature = "serde",
-    derive(serde::Deserialize, serde::Serialize),
-    serde(bound(serialize = "E: serde::Serialize")),
-    serde(bound(deserialize = "E: for <'d> serde::Deserialize<'d>"))
-)]
pub enum RaftError<NID, E = Infallible>
where NID: NodeId
{
    #[error(transparent)]
    APIError(E),

+  #[serde(bound = "")]
    #[error(transparent)]
    Fatal(#[from] Fatal<NID>),
}

[drmingdrmer]

Looks good. Let me check it in this PR:

• Refactor: RaftError: serde derive #993