[kvlonge]

Why

Snowflake metadata row counts ran one information_schema.tables query per relation, and each one needs a running warehouse. A code location sampling statistics therefore kept the warehouse busy for about a second per relation. SHOW OBJECTS returns the same rows and bytes from metadata alone, so no warehouse has to run.

Changes

• estimate_relation_row_count on Snowflake reads SHOW OBJECTS LIKE '<name>' IN SCHEMA <database>.<schema> and matches the stored name exactly. Kind TABLE reports rows and bytes, with basis show_objects_bytes. Views report no count, as before.
• Two cases keep the bound-parameter information-schema query, matching the adapter's other SHOW lookups: a relation without a schema, and a name containing a backslash.
• The unit tests cover tables, views, a neighbour matched by the LIKE wildcard, the backslash name and the schema-less read.

DuckShop edge case: not applicable. This changes how one adapter reads metadata, not a project shape, data value or lifecycle.

Verification

• pytest tests/unit/src/sqlbuild/adapters/test_relation_row_count_estimate.py tests/unit/src/sqlbuild/adapters/snowflake: passes. The statistics publication integration tests pass.
• Live Snowflake check, X-Small warehouse, 63 relations: SHOW OBJECTS matched information_schema.tables exactly for every relation (28 tables with rows and bytes, 35 views with no count).
• A code location sync with all 63 statistics due completed in 44 s. It ran no per-table warehouse queries: 63 SHOW calls at about 0.1 s each, where each per-table query used to take about 1 s.
• make check-ci: passes.