[MAINTENANCE] Add integration coverage for quoted schema names on PostgreSQL

[joshua-stauffer]

Summary

Wrapping an identifier in the dialect's quote characters is how a user asks GX to use that identifier verbatim rather than case-folding it. table_name honors that convention: a quoted value is unwrapped and rebuilt as a sqlalchemy.quoted_name(..., quote=True), so the quotes govern the emitted SQL instead of becoming part of the name.

A schema name never gets the same treatment. _effective_schema_name returns the string with the quote characters still embedded, so SQLAlchemy escapes them as data rather than honoring them as quoting:

sa.table("mytable", schema='"MySchema"')
# FROM """MySchema""".mytable          <- targets a schema whose name contains quotes

sa.table("mytable", schema=sa.quoted_name("MySchema", quote=True))
# FROM "MySchema".mytable              <- what was asked for

The practical consequence is that a case-sensitive schema is unreachable: quoting is the only way to preserve its case, and quoting is what breaks. The same mismatch also makes the schema-existence check in TableAsset.test_connection compare a quoted name against a bare schema listing, so it reports a schema that does exist as missing.

This PR adds coverage only — no behavior change.

What's here

Two PostgreSQL integration tests in tests/integration/data_sources_and_expectations/test_quoted_schema_names.py that differ in exactly one variable: whether the schema is handed to the asset bare or bracketed by quotes.

• test_bare_schema_name_reaches_the_table — passes. A control, so the companion test's failure can only be attributable to the quoting rather than to anything about naming a schema on the asset.
• test_quoted_schema_name_reaches_the_table — xfail(strict=True). Documents the current behavior and converts into a hard failure the moment the identifier handling is corrected, so the fix cannot land without this test being updated deliberately.

The shared PostgreSQL setup targets its schema through the connection string's search_path, which never passes through GX's identifier handling, so these tests name the schema on the asset instead. That is also why PostgreSQL is the backend here: it runs locally, and the schema is created and dropped by the existing harness.

Verification

$ pytest tests/integration/data_sources_and_expectations/test_quoted_schema_names.py -m postgresql
1 passed, 1 xfailed

Against a local PostgreSQL container. With --runxfail, the xfailed test fails as described above:

TestConnectionError: Attempt to connect to table: ""gx_ci_test_17bef88060".expectation_test_table_..."
failed because the schema ""gx_ci_test_17bef88060"" does not exist.

ruff check, ruff format --check, and the repo's CI type-check configuration are all clean.

Notes for review

• On PostgreSQL the only route to this code path is the asset's schema_name, which is deprecated but still supported. Backends that carry a schema in structured connection details (Snowflake, SQL Server) reach the same _effective_schema_name through their non-deprecated schema_ property, so the defect is not confined to the deprecated argument.
• A fix would most likely normalize the schema at _effective_schema_name — the single point both routes pass through — mirroring what the table_name validator already does, including the dict() round-trip that re-adds the quote characters so stored configs keep their shape.

[netlify]

✅ Deploy Preview for niobium-lead-7998 canceled.

Name	Link
🔨 Latest commit	32459e0
🔍 Latest deploy log	https://app.netlify.com/projects/niobium-lead-7998/deploys/6a7a1efd4e41ce000860c3f7