[jat255]

Closes #447

This PR makes R cast a Snowflake semantic view's variables to the type the view declares. Snowflake refuses a value of another type, and Snowflake ODBC driver 4.x binds a number as text, so a parameterized metric call failed.

Details (agent-written)

semantic_parameters_from_spec() now keeps a Snowflake variable's declared type. A new helper in catalog-snowflake.R writes VARIABLES "MIN_TOTAL" => CAST(? AS NUMBER(10,2)), and both the metric query and the access probe use it. The type is put into SQL, so only a plain type name such as NUMBER(10,2) or TIMESTAMP_TZ is kept. Any other declaration leaves its variable with a bare placeholder, as before.

I cast to the declared type rather than to a fixed one: CAST(? AS NUMBER) also passes Snowflake's check, but it rounds 50.5 to 51.

Testing: the full R suite passes, with new cases for the cast and for a declared type that is not a plain name. On live Snowflake, every R Snowflake live test passes with this stack, including the three semantic view tests. I ran it with driver 4.0.0 on macOS only. CI on Linux may bind numbers differently, but the cast is correct for either.

Kata: whhh.

R changes

A Snowflake semantic view variable is now bound with a cast to the type the view declares (catalog-snowflake.R). A call such as call_metrics(..., arguments = '{"MIN_TOTAL": 50}') against a view that declares MIN_TOTAL NUMBER(10,2) used to fail with a type mismatch, and now returns the metric. Typed arguments for Snowflake gain an internal sql_type field. Databricks metric views are unchanged. NEWS has a bullet.