feat(schema): unblock fold_db_node openapi.rs registration ( path-prefix + FieldBase inline)

[shiba4life]

Summary

Phase 4 follow-on to slice 3 (#681) and slice 3-extension (#682). The slice-3 family registration in fold_db_node hit four classes of utoipa $ref resolution issues post-cascade:

1. $ref path-prefix quirks

utoipa renders fully-qualified type paths with dots:

```
"crate::schema::types::field_value_type::FieldValueType"
→ "crate.schema.types.field_value_type.FieldValueType"
```

which don't match the registered component names. Fixed via module-scope `use` statements in:

• `access/types.rs` — `super::capability::{CapabilityConstraint, CapabilityKind}`
• `schema/types/declarative_schemas.rs` — `FieldAccessPolicy` + `FieldValueType`
• `schema/types/field/base.rs` — `atom::{Molecule, MoleculeHash, MoleculeRange, MoleculeHashRange}`

so the bare type names appear in field declarations / alias macros.

2. `#[aliases]` doesn't substitute into bare `FieldBase` references

utoipa's `#[aliases(...)]` on `FieldBase` registers aliases (`FieldBaseSingle`, `FieldBaseHash`, etc.) as separate components but does NOT substitute alias names into bare `FieldBase` references in dependents (`SingleField`/`HashField`/etc. via `#[serde(flatten)]`).

Fixed by adding `#[schema(inline)]` to each variant's `base` field so the FieldBase shape is inlined directly into the variant schema rather than $ref'd. Avoids the larger inline-FieldBase refactor that would have required moving 60+ `.base.X` accessors across the field/ module and `variant.rs`.

3. `SchemaWithState.schema` referenced the `Schema` alias

`SchemaWithState` had `pub schema: Schema` where `Schema` is a `pub use crate::schema::types::declarative_schemas::DeclarativeSchemaDefinition as Schema;` re-export. utoipa emits `$ref: Schema` (alias name), but the canonical component is named `DeclarativeSchemaDefinition`.

Fixed by changing the field type to `DeclarativeSchemaDefinition` directly. Drops the alias usage in this load-bearing wire shape.

Why

Without these fixes, fold_db_node's Phase 4d' regen produces 14 unresolved $refs even though all the underlying types have ToSchema. After this lands + the bump cascade, fold_db_node's openapi.rs registration resolves cleanly with zero unresolved $refs.

Parent: gbrain `projects/api-typegen-unification`. Recipe: gbrain `projects/register-schema-field-bodies`.

Test plan

• `cargo fmt --all` clean
• `cargo clippy --workspace --all-targets -- -D warnings` clean
• `cargo test --workspace --all-targets` — 662 core tests + observability all green
• CI green
• Bump cascade picks this up

🤖 Generated with Claude Code