feat(typespec): close datapackage.json -> validate loop (item 8l)

[discreteds]

Summary

Closes item 8l (backlog: mountainash-central/01.principles/mountainash/h.backlog/active/frictionless-validation-loop-closure.md) — the vision's headline Phase-1 loop: datapackage.json → DataResource → TypeSpec → contract → validate.

• DataResource.to_typespec() — passthrough (already-TypeSpec) / convert (raw dict via typespec_from_frictionless) / raise (garbage type). table_schema is None returns None (legitimately optional).
• DataResource.to_contract() — thin delegation to TypeSpec.to_contract(), raises ValueError when there's no schema to compile.
• End-to-end cross-backend test (tests/relations/dag/cross_backend/test_datapackage_validation_loop.py): a real on-disk datapackage.json (loaded via the Path branch of DataPackage.from_descriptor, not the dict branch every other test uses) with two resources (primaryKey, pattern, range, foreignKeys, enum), proving dag.validate(specs, backend=...) on polars/narwhals/ibis:
  • conforming data passes on all 3 backends
  • a duplicate primary_key raises IdentityInvalidError on all 3 (confirms item 8j's raise-by-default characterization holds for a descriptor-sourced TypeSpec, not just hand-built ones)
  • drifting data reports specific failing check IDs (score__le, status__isin, the FK anti-join) on all 3 backends
• Vision doc fix (mountainash-central, committed directly to main per that repo's convention): unique=True row was stale (n_unique().eq(len())), corrected to the shipped row-level is_duplicated().not_(); added a footnote documenting the null-precondition guard on nullable range/pattern/enum checks.

Scope note — cross-backend pattern check narrowed to Polars

The drifting-data test's cross-backend pattern-constraint assertion surfaced a genuine pre-existing defect, unrelated to this item's to_typespec()/to_contract() seam:

• Narwhals regexp_match_substring is a complete no-op stub (return input — never evaluates the pattern).
• ibis-duckdb (the default ibis.get_backend() dialect) returns '' (not NULL) from re_extract on no-match, so compile_datacontract's .is_not_null() check is always True.

Both make the pattern check silently report "passed" for values that violate it. Range/enum/FK checks are confirmed correct on all 3 backends — this is isolated specifically to pattern. Per the plan's explicit I2 disposition (any pre-existing cross-backend defect discovered here is out of scope — compile_datacontract/dag.validate are already-shipped, already-tested code this item does not touch): narrowed the pattern assertion to a dedicated Polars-only test (test_pattern_violation_detected_polars), and filed backlog item 103 with full probe evidence and candidate fix shapes (mountainash-central/01.principles/mountainash/h.backlog/active/pattern-constraint-check-silently-passes-narwhals-ibis-duckdb.md).

Verification

• hatch run test:test-quick — 22991 passed, 0 failed, 81 skipped, 3465 xfailed (full suite).
• hatch run ruff:check — clean.
• hatch run mypy:check — zero new errors (pre-existing import-untyped baseline noise only; none touch the changed files).

Spec / Plan

• Spec: mountainash-central/04.planning/mountainash/superpowers/specs/2026-08-17-frictionless-validation-loop-closure-design.md (rev 3)
• Plan: mountainash-central/04.planning/mountainash/superpowers/plans/2026-08-17-frictionless-validation-loop-closure.md