refactor(hyperlane): route field accessors through Data macros

[vshengbro]

Summary

Adds multipart/form-data and URL-encoded body parsing to http-type, and
replaces every hand-written struct field accessor in the workspace with
lombok Data-generated accessors so field access has one source of truth.

Two request defects surfaced while wiring up the new body API and are
fixed here: repeated query keys silently lost every value but the last,
and header lookup failed for any key that was not already lowercase.

Changes

Body parsing (new):

• type/src/body/struct.rs — BodyPart and MultipartFormData
• type/src/body/impl.rs — multipart and URL-encoded parsing, plus
  part-level and form-value access
• type/src/body/const.rs, type/src/body/type.rs — internal constants
  and the BodyFormValue / content-type aliases
• type/tests/body/ — multipart and form tests

Request correctness:

• type/src/request/type.rs — RequestQuerysValue becomes
  VecDeque<String> so a repeated key keeps all its values
• type/src/request/impl.rs — query parser and accessors handle the
  multi-value case; header lookup is case-insensitive to match the
  lowercased keys the header parser stores
• type/tests/request/fn.rs — multi-value query coverage

Accessors and visibility (workspace-wide):

• 21 hand-written get_* / set_* accessors removed in favour of
  Data derives across RequestConfig, HttpRequest, WebSocket,
  WebSocketConfig, Body and RequestBuilder
• request/src/request/config/impl.rs deleted outright once its accessors
  became generated
• business code no longer reads struct fields directly, going through
  accessors instead
• field visibility narrowed to the least each field needs: pub(super)
  inside a module tree, pub(crate) across sibling modules, pub only
  for cross-crate and integration-test consumers
• RequestBuilder::request keeps its public mutable accessor because
  request/tests/builder consumes it

Verification

• cargo test --workspace — 724 passed, 0 failed
• cargo clippy --all-targets — 0 warnings
• crate fmt --check — clean, idempotent
• audit_rust_standards.py . — 51/51 PASS
• Independent scan for hand-written field accessors — 21 before,
  0 after
• git commit ran the real pre-commit gate (no --no-verify);
  0 new violations
• Commit author is the global-config identity, vshengbro <root@ltpp.vip>

Notes

WebSocketConfig deliberately keeps its hand-written set_* methods.
They are fluent builders taking mut self and returning Self, not
field setters, so converting them to macro-generated setters would change
their meaning. Those fields are marked #[set(skip)]; their getters are
generated like everything else.

type(clone) is required rather than redundant on the 13 Option<T> and
Method fields that carry it. Checked against lombok 10.2.2 source:
ReturnType::Reference is the default, so a Clone field without the
attribute yields &T; and for Option<T> the default branch generates
self.field.clone().unwrap(), which would turn "unset" into a panic.
Removing the attribute from each of four representative fields was tried
and all four failed to compile, so it stays.

WebSocketConfig holds &mut references and therefore cannot derive
Clone, Debug or Default; it was previously carrying no derives at
all and nothing in the workspace depends on those traits for it.

Parser failures return Option, following the existing try_get_*
convention, rather than a bespoke error type.

🤖 Generated with Hermes