[kvlonge]

Why

Scenarios and macro-test scope scans still fell back to a Polyglot re-parse when the CTE scanner failed. That accepted quoted CTE names and AS [NOT] MATERIALIZED, rewrote the bodies, and kept a Python-only path. SQL tests already reject these forms (#1026).

Accepted-input changes (no external users, so no shim; no project in the repo or differential corpus uses these forms):

• Quoted scenario and macro-test CTE names are errors.
• AS MATERIALIZED / AS NOT MATERIALIZED are errors in SQL tests, scenarios and macro tests.
• Unquoted non-ASCII scenario CTE names (e.g. café), previously accepted, are errors (ASCII-only, as for SQL tests).
• CTE names starting with a digit (e.g. 2024_orders) get the located name error with help.
• A scenario's located CTE-name error is reported before its other extraction errors, matching SQL tests.

Changes

• Removed the Polyglot CTE fallback (analysis/ctes.py) from scenario extraction, unclassified SQL-test CTEs and macro-test relationship scans. The native scenario outcome has one error.
• Scenarios are validated before compile like SQL test blocks, with a located error (path:line:column) and the SQL-test help.
• Materialization hints get a located error naming the CTE instead of the misleading must use AS (...).
• Discovery records each scenario's SQL span for locations.
• testing.mdx and scenarios.mdx document the rules; the skill reference is regenerated.

Verification

• CLI E2E on python, native and native-preview for a quoted scenario CTE, a materialized scenario CTE and a quoted macro-test CTE: identical located errors with help. Cases are in the failure corpus.
• waffle_shop compiles to identical artifacts before and after on all engines.
• Rust tests (incl. scenario body span with CRLF and non-ASCII), clippy/fmt on 1.99.0, focused tests, compiler differentials, make check-ci, hygiene.