[kvlonge]

Why

SQL test files needed ceremony the compiler could work out: a TEST(); header in every file, an explicit mode that the CTE names already determine, and a trailing SELECT 1 that is ignored.

Changes

• A single-test file under tests/unit may omit its TEST(); header; files with several blocks still need one per block.
• TEST mode is inferred from a top-level __macro_actual__, __udf_actual__ or __table_fn_actual__ CTE; an explicit mode is never overridden, and a contradicting one is still an error.
• Tests and scenarios may end after their last CTE. Compile, Rules and the formatter add the omitted SELECT 1 internally through one shared helper; the formatter never writes it back.
• The omitted-SELECT 1 check returns early for bodies that end in SELECT 1 and only scans quote, comment and bracket positions otherwise, with a time-limited guard test.
• Testing, scenarios, macros and visibility docs show the short forms, keeping one explicit example; waffle-shop uses both forms.

Verification

• Unit, integration and E2E tests for header-less files, mode inference, omitted SELECT 1, formatter and Rules handling; a native extraction test for the optional SELECT 1.
• Compared with 0.134.8 on a large real project: identical diagnostics, Rules findings, format results, compiled SQL and manifest; compile CPU and memory at parity. Stripping the optional forms from every test file gave identical compiled output.
• Focused pytest, ruff, fensu check and make check-ci.