fix(model): H2 quoting follow-ups from #4628 (migrator DDL, tenant JOIN memo key, classifier coverage) #4629

Follow-ups from the independent review of #4628 (H2 identifier quoting, closed issue #4440). None blocks the merged fix; each is a separate, scoped change.

1. H2Migrator DDL quoting. H2Migrator.quoteTableName()/quoteColumnName() (vendor/wheels/migrator/H2Migrator.cfc:93-101) still emit unquoted names, so a migration cannot create a reserved-word column on H2 even though the model layer can now query one. PostgreSQL, Oracle and CockroachDB migrators have the same gap, so this wants one decision covering all folded-DDL adapters rather than an H2-only patch.

2. Association JOIN memo key and datasource-dependent quoting. sql.cfm:2370 memoises the JOIN fragment under sd{n}_alias{n} in application-scoped class data. That fragment now contains H2 quoting that depends on the effective datasource's fold case, so two H2 tenant datasources with different DATABASE_TO_UPPER/DATABASE_TO_LOWER settings could be served each other's memoised JOIN. It fails loudly (wrong-case identifier not found), but the key should include the fold case or effective datasource for H2.

3. Classifier unit coverage. $readIdentifierCase() (vendor/wheels/databaseAdapters/H2/H2Model.cfc:102-112) has no direct test: the unit tests inject identifierCase through H2CaseProbe and bypass the classifier, and the only real assertion is "upper" on the suite datasource. Extracting a pure $classifyIdentifierCase(settings) and table-testing the verified shapes (2.x default, 2.x DATABASE_TO_UPPER=FALSE, 2.x DATABASE_TO_LOWER=TRUE, 1.3.x with no DATABASE_TO_LOWER) would pin it. Related: add the 1.3 evidence to the comment at H2Model.cfc:102-105 — the "no DATABASE_TO_LOWER ⇒ upper" shortcut is only correct because H2 1.3 compares quoted identifiers case-insensitively, and without that note it reads like a bug.

4. Spec fixture hygiene. vendor/wheels/tests/specs/model/OrderReservedWordSpec.cfc:109 runs DELETE FROM c_o_r_e_casepreservation with no WHERE. Harmless today (nothing seeds that fixture), but scope it to the ids the spec creates if the fixture ever gains data.

Evidence: review at task wheels-7; H2 semantics verified against H2 1.3.172, 2.1.214, 2.2.224 and 2.3.232.