Support renaming in zlink-macros derives

[zeenix]

Fixes #81.

Renaming is now supported in the derives, driven by zlink's own attributes:

• #[zlink(rename = "...")] on a field or variant sets its name.
• #[zlink(rename_all = "...")] on a struct applies a case convention to its fields, on an enum to
  its variant names, and on a variant to that variant's fields. An explicit rename overrides it.
• On CustomType, a container rename names the IDL type itself.
• ReplyError variants can be renamed, changing the serialized error name.

The eight rename_all values and their semantics are serde's (lowercase, UPPERCASE,
PascalCase, camelCase, snake_case, SCREAMING_SNAKE_CASE, kebab-case,
SCREAMING-KEBAB-CASE) — the attributes are ours, but there's no value in inventing different
semantics for something Rust users already know.

So the workaround posted on the issue becomes:

#[derive(Debug, Serialize, Type)]
#[serde(rename_all = "camelCase")]
#[zlink(rename_all = "camelCase")]
struct Membership {
    user_name: String,
    group_name: String,
}

On serde's attributes

The issue asked to investigate reusing serde's attributes. That was investigated and rejected:
it's consistent with the existing #[zlink(rename)] API on proxy methods and ReplyError fields,
and it works on types that don't derive serde at all — the ReplyError derive writes its own serde
impls, so #[serde(..)] wouldn't even compile there.

The trade-off is real and accepted: a type deriving both serde's and zlink's traits must state the
renaming twice, and nothing detects a disagreement. The derive docs say so plainly rather than
leave it to be discovered.

Two design points worth review

ReplyError variant renaming changes both derives in a single commit. The serde derive puts
the name on the wire, the introspection derive puts it in the IDL; both resolve through one shared
naming module, because a disagreement between them would be an interface description that lies
about the protocol.

A container rename is a hard error on the Type derive (its object type is anonymous) and on
both ReplyError derives (error names are qualified by #[zlink(interface)]). Silently ignoring
it is how people ship wrong IDL. This is technically breaking for anyone who wrote a rename that
never did anything.

Also here

The first commit is a standalone fix for a pre-existing bug: #[zlink(..)] parsing was
order-dependent in both directions — #[zlink(borrow, rename = "x")] silently dropped the rename,
and #[zlink(rename = "x", borrow)] silently dropped the borrow (losing zero-copy). The rename
work builds on these helpers and needs them to be order-independent.

Not included

• #[serde(skip)]/flatten/tag equivalents. These also affect what's on the wire and so can
  produce wrong IDL, but they aren't renaming. Introspection derives describe fields serde doesn't send: no equivalent for skip, flatten, tag #291
• Raw identifiers remain as they are today: a r#type field fails to compile in the introspection
  derives and reaches the wire as the r#type key. Pre-existing, and fixing it would change an
  existing derive's wire format. Raw identifiers are mishandled by the derives: panic in introspection, r# leaks to the wire #290

Testing

cargo test --all-features: 458 passed, 0 failed (429 on main). The no_std path
(cargo test -p zlink-core --no-default-features --features idl-parse,proxy,defmt): 154 passed.
Clippy -D warnings and nightly fmt clean.

New tests cover all eight conventions for both fields and variants, rename overriding
rename_all, kebab-case names not breaking generated statics, the CustomType name and its
Type::Custom reference staying in sync, and — the important one — that the IDL error/field names
match what actually goes on the wire.

Generated by Claude Opus 4.8 (1M context).

[codspeed-hq]

Merging this PR will not alter performance

✅ 4 untouched benchmarks

---

_{Comparing claude/github-issue-81-rename (03bfe7d) with main (0ab98f1)}