feat(codegen-ts)!: derive native uuid[]/text[] from isArray; drop uuid_array/text_array dbColumnType values (phase 1)

[dmealing]

dbColumnType slim-and-derive — Phase 1 (TypeScript + migrate-ts)

Applies the project's derive-don't-declare principle (ADR-0023) to the @dbColumnType escape hatch. The array-ness of a column already lives on the field's isArray axis, so native Postgres arrays are derived, not declared. Breaking metamodel-vocabulary change, scoped to the pre-1.0 consolidation window.

Per the design spec (docs/superpowers/specs/2026-06-29-dbcolumntype-slim-and-derive-design.md), this PR is only the TS port + migrate-ts. Other ports and the shared fixtures/ corpora are coordinated separately.

The change

• Vocab (metadata/db-constants.ts): removed uuid_array + text_array from DB_COLUMN_TYPE_VALUES + DB_COLUMN_TYPE_LEGAL_SUBTYPES. The loader now rejects both (ERR_BAD_ATTR_VALUE). Constants stay exported. Kept uuid / jsonb / timestamp_with_tz (the timestamp_with_tz default-flip is Phase 2).
• migrate-ts (canonical DDL): subtypeToSqlType now derives native arrays from isArray — field.string isArray → text[], field.uuid isArray → uuid[] — via a small arrayElementSqlType helper; dropped the retired uuid_array/text_array overrides. Scalar field.uuid → uuid, no-maxLength field.string → text (unchanged).
• Description text left unchanged on purpose. The @dbColumnType attr description is carried verbatim in two shared cross-port fixtures (fixtures/registry-conformance/expected-registry.json, fixtures/metamodel-docs/expected/). Those are coordinated separately, so this PR does not edit the description (the loader-level value-set enforcement is in db-constants.ts, independent of the doc text). A follow-up alongside the other ports can refresh the description + those fixtures.

What TS already did vs the field.uuid[] question

• field.uuid scalar was already handled → uuid() column (not a gap, contrary to the worst case the brief flagged).
• field.uuid + isArray already emits uuid() + .array() (= uuid[]) via the column-mapper's scalar-array path (uuid is neither object nor map), and field.string + isArray already emits text[]. No column-mapper change was needed — added explicit test coverage to lock it in.
• field.string no-maxLength → text, with maxLength → varchar(N): already correct, confirmed.
• The only real fix was in migrate-ts, where native arrays were only reachable through the now-removed *_array overrides — isArray was never consulted. Now derived.

Test evidence

• Targeted unit suites green: metadata attr-schema-validate (incl. new uuid_array/text_array → ERR_BAD_ATTR_VALUE cases) + db-definition-completeness; codegen-ts column-mapper (new field.uuid/field.string isArray → uuid[]/text[]); migrate-ts expected-schema (derived arrays, scalar uuid/text, removed-value rejection). Plus the registry-conformance + metamodel-docs byte-gates stay green (description left unchanged). 131/131 across those 6 suites.
• typecheck clean for metadata / codegen-ts / migrate-ts; pre-push CI-parity build+typecheck gate passed.
• Canonical schema.postgres.sql is unchanged — no persistence fixture exercises native arrays, so removing the overrides produces no diff in the committed canonical schema.

Part of dbColumnType slim-and-derive Phase 1.

🤖 Generated with Claude Code

[dmealing]

Folded into the integrated Phase-1 PR (feat/dbcolumntype-slim-derive) which merges all five ports atomically — that's the one that lands. Keeping this open for per-port review history.

[dmealing]

Landed via the integrated Phase-1 PR #121 (squash-merged to main).