DatasetMetadata: express reused-term class linkage with schema:domainIncludes (fixes #92)

[Haigutus]

Fixes #92 — implements the schema:domainIncludes convention discussed in that thread.

What

In NCP/RDFS/DatasetMetadata-AP-Voc-RDFS2020.rdf, the class↔attribute linkage of reused external terms (dcterms:, prov:, dcat:, adms:, RDF reification) is now expressed with the non-inferential schema:domainIncludes instead of rdfs:domain; CIM-owned terms (dcatcim:, Metadata-European:) keep rdfs:domain.

This:

• removes the ontology hijacking of rdfs:domain on external terms (it entails every title-bearing resource is a dcat:Dataset), which @VladimirAlexiev correctly flagged;
• keeps the class↔attribute structure in the RDFS vocabulary (not only in SHACL), so RDFS-driven schema / Simple-SHACL generation still has something to derive from;
• restores the linkage lost in 489c5c5 for the 11 URI-typed Dataset attributes (8 dcterms: + prov:wasGeneratedBy via schema:domainIncludes; Metadata-European:processType / usedSettings via rdfs:domain), and applies the convention consistently to the string-typed reused terms too.

Verification

• cims:dataType and cims:multiplicity are unchanged — no drift vs ncp-v2-4-1.
• The URI/literal classification was cross-checked against the owning vocabularies (DCMI Terms, PROV-O): every #URI attribute has a class-valued range (RightsStatement, Frequency, Standard, LicenseDocument, Agent, Location, prov:Activity) and all literals match. (Separately: dcterms:rights / dcterms:rightsHolder are modelled as #String here although DCT ranges them as classes — pre-existing, unchanged, out of scope for this PR.)
• The file parses; multi-domain terms became multiple schema:domainIncludes (a union — the intended "used on either" — rather than the conjunction that multiple rdfs:domain implies).

Docs — ⚠️ breaking for rdfs:domain-driven tooling

.github/README.md (§ RDF schemas) now states the RDFS ↔ SHACL division of responsibility and the RDFS conventions explicitly: flat RDF/XML (no nested nodes); RDFS holds classes + attributes + simple attribute-local constraints (datatype, multiplicity) + the owned/reused class linkage; Simple SHACL is a mechanical projection of the RDFS; Complex/Validation SHACL carries what RDFS cannot express. Tooling that derived attribute-to-class membership from rdfs:domain must now also read schema:domainIncludes for external terms.

@VladimirAlexiev — this implements the non-inferential-domain approach from #92; would appreciate your review.

[VladimirAlexiev]

@griddigit-ci @Sveino such fix needs to be done in UML->Ontology, right?

I like domainIncludes for its lightness. But it can't express cardinality.
And @Haigutus in some circumstances cardinality can he considered part of the model. Eg in GraphQL a scalar (single valued) is different from an array (multi valued).
But descriptive data props are always multivalued because they can be expressed in multiple languages. Then you need sh:uniqueLanguage.

Oh well... in the balance, i think that domainIncludes is useful.

[griddigit-ci]

@VladimirAlexiev we are not really able to do this thing in UML due to tooling. So it seems that this RDFS is somehow crafted manually/semi manually
I see the changes from Kristjan and they are fine with me so I will approve the pull request. The in some of the documents we will need to describe what we are doing here

[VladimirAlexiev]

dcat: props can stay with domain not domainIncludes