…sion-column checks

BigQuery rejects "cast(... as varchar)"; use the cross-adapter type
macro so check_extension_columns_in_core_{eligibility,medical_claim,
member_months,pharmacy_claim} tests compile on every warehouse.

Collapse the repeated per-column cross-adapter jinja
(bigquery/databricks string, athena/databricks real, fabric
datetime2, etc.) into three shared anchors — *string, *datetime,
*float — and reference them from every column_types entry.

Also gate each synthetic_data__* seed on use_synthetic_data so
non-synthetic runs stop materializing seeds nothing refs.

Drops ~260 lines of repetition from seeds/synthetic_data/synthetic_data_seeds.yml
without changing runtime type resolution.

… at compile time

Returns ref('raw_data__<table>') when use_synthetic_data is true, or
source('source_input', <table>) otherwise. Because ref() is evaluated
at parse time, dbt wires the seed as an upstream dependency of every
bridge model that calls tuva_source(), so seeds run before their
dependent models without an on-run-start hook or explicit ordering.

Returns the Relation object (not a rendered string), so callers can
use either "from {{ tuva_source('X') }}" or bind it with {% set r =
tuva_source('X') %} for adapter.get_columns_in_relation(r).

… into single SELECT via tuva_source()

Every bridge model had two duplicated SELECTs gated by
use_synthetic_data. Replaced with one SELECT reading from
tuva_source('<table>') — the macro swaps ref() vs source() at
compile time so the toggle is invisible to the model.

With _sources.yml restored, setting input_database + input_schema
now points the same models at the user's own input layer when
use_synthetic_data is false. No model edits needed.

…eprecation

dbt now flags custom top-level config keys; nest the seed
batch_size hint under +meta so the deprecation warning goes away.

dbt 1.10+ requires generic-test arguments to be nested under
arguments (MissingArgumentsPropertyInGenericTestDeprecation).
Updated the hcc_recapture staging/intermediate/final yml tests.