Usage metaclass classification is conflating metaclass with referential/composite syntax signals

[groundhog-tech]

The current usage-metaclass classification appears to mix up two different dimensions:

usage metaclass / kind (PartUsage, ReferenceUsage, ...)
referential/composite semantics
These are not the same thing.

Per SysML v2 §7.6.4, ReferenceUsage is a usage kind. It is not the generic semantic representation of “referential usage”.

The ref keyword should contribute to referential/composite semantics, not be used as the source of usage metaclass classification.

Minimal example：
package sample {
part a;
part b {
out a;
ref part c;
}
}
Current syster result：
sample::b::a -> PartUsage with direction = out
sample::b::c -> ReferenceUsage

Why this looks wrong
ref part c should not become ReferenceUsage just because ref appears.The ref keyword is a signal for referential/composite semantics, not for deciding whether the usage metaclass is ReferenceUsage.

Expected result：
sample::b::a -> ReferenceUsage with direction = out
sample::b::c -> PartUsage with isReference = true

Likewise, usages without an explicit kind keyword should not simply fall back to PartUsage by default.

Likely cause
The current classification pipeline seems to do exactly that:

base/src/parser/ast/elements.rs
Usage::usage_kind() includes REF_KW => Ref
base/src/parser/ast/mod.rs
token_to_enum_method! scans the whole Usage subtree
base/src/hir/symbols/helpers.rs
UsageKind::Ref -> InternalUsageKind::Reference
None -> InternalUsageKind::Part
So the metaclass classifier is currently consuming syntax that should belong to referential/composite semantics instead.