[kvlonge]

Why

sqb format re-indents expression macro bodies using SQLBuild's lexical rules for the project's dialect. Databricks and DuckDB shared one entry with no backslash escapes, no raw strings and flat block comments. SQLQuack and both adapters read them differently: Databricks honours \' in '...' and "..." outside r'...', DuckDB honours it inside E'...', and both nest block comments. A multi-line string after \' therefore looked closed early, and its next line lost two spaces on every run, changing the value (6 → 4 → 2 → 0 spaces) so sqb format --check never settled.

Changes

• Databricks gets its own entry: backslash escapes in ' and ", the r'...' raw prefix and nested block comments.
• DuckDB falls through to the default PostgreSQL entry (E'...' escapes, nested block comments), as its adapter already declares.
• The same rules decide quoted text for semantic-validation normalisation, fast lineage, dynamic pivot, SQL tree nodes and query-facts shapes; markers after such strings or nested comments are now found where they were skipped.
• Tests: scanner cases for both dialects, including raw strings, a plain DuckDB string and nested comments; format-twice cases on Databricks, DuckDB E'' and a nested comment holding an apostrophe; the CLI format test now covers Databricks and DuckDB projects.
• DuckShop is a Snowflake project and cannot express a Databricks- or DuckDB-only lexical case; its Snowflake backslash case already exists.

Verification

• The new scanner, format and CLI cases fail before the fix and pass after it.
• sqb format three times on the repro macro: before, kept indent loses two spaces per run on both dialects; after, it keeps all six on every run.
• cargo test -p sqlbuild-sqltext -p sqlbuild-rules -p sqlbuild-analysis --lib, make test and make check-ci pass.