test: re-enable base fixture validation; harden AAC tier validator

[larrybabb]

Addresses two review findings from #78. Targets the ballot-updates branch so the fixes land in that PR.

1. Base-namespace fixtures were silently unvalidated

The va-spec 1.1.0-ballot.2026-09 submodule labels core/base test fixtures with the namespace va-spec (no dot); profile fixtures use va-spec.<profile>. The fixture-discovery loop in test_va_spec_fixtures_validation.py only kept namespaces matching startswith("va-spec."), so all 39 base fixtures were dropped and test_va_spec_fixtures validated only the 20 profile fixtures — while still passing green, hiding the loss of coverage.

Fix: accept the bare va-spec namespace (routed to VaSpecSchema.BASE) and continue past any namespace get_va_spec_schema doesn't recognize (also removes a latent KeyError on VA_SPEC_TEST_DEFINITIONS[None]).

Routing after the fix (verified against the pinned submodule's test_definitions.yaml):

schema	fixtures
BASE	39 (was 0)
AAC_2017	11
ACMG_2015	5
CCV_2022	4
skipped (vrs)	2

2. AAC validate_tier_evidence_lines crashed on non-dict input

VariantClinicalSignificanceStatement.validate_tier_evidence_lines runs in pydantic mode="before" and called values.get(...) unconditionally, so validating from a non-dict input (e.g. model_validate("propositions.json#/1")) raised AttributeError instead of a normal validation error. Added an isinstance(values, dict) passthrough guard, matching the existing nested type guards in the same validator.

Verification

• ruff / ruff-format pre-commit hooks pass on both files.
• Fixture routing table above reproduced with the real submodule namespaces.
• Non-dict guard exercised (str / list / None pass through; dict still processed).

Note: the local environment has stale dependency versions (vrs 2.0.0 / cat_vrs 0.5.0, no ga4gh.core.metadata), so the re-enabled base fixtures could not be executed here — CI on the updated deps should confirm they validate.

🤖 Generated with Claude Code