Fix/grouping sets comma (#6645)

• chore: clean checkstyle violations in SQLServerCollateTest

Trim the blank line after the class opening brace and the trailing blank lines at end-of-file so the maven-checkstyle-plugin (bound to the validate phase) stops blocking the surefire run. Pure formatting fix, no logic change.

• chore: import openspec scaffold (config, templates, baseline specs)

Imports the openspec workflow scaffolding that lives on the sibling branches but has not yet landed on master:

openspec/config.yaml — project context + per-artifact rules openspec/templates/ — proposal/spec/tasks/design templates openspec/specs// — baseline spec.md for 8 capabilities (connection-pool-core, dialect-registration, feature-gate-naming, filter-chain, monitoring-stat, sql-parser-core, test-infrastructure-core, wall-security)

Sourced verbatim from the existing branches that introduced the layout; no content changes. This commit is content-only; the change-record / spec delta for the GROUPING SETS fix lives in a follow-up commit.

• fix: preserve comma separator before GROUPING SETS in GROUP BY

ODPS / Hive / Oracle distinguish two surface forms in GROUP BY:

GROUP BY a, GROUPING SETS((a, b), (a), (b), ())   -- with comma
GROUP BY a GROUPING SETS((a, b), (a), (b), ())    -- no comma

Before this fix the parser collapsed both into the same AST and the output visitor unconditionally dropped the comma, so a parse + emit roundtrip silently rewrote the SQL.

Three coordinated changes:

• SQLGroupingSetExpr gains a hasPrefixComma flag (default true to preserve the historically dominant emit shape for AST builders that did not previously distinguish the two forms). The flag participates in clone(), equals(Object) and hashCode().

• SQLSelectParser.parseGroupBy tracks whether the previous loop iteration consumed a comma; when the next parsed item is a SQLGroupingSetExpr and no comma was consumed, the parser marks it hasPrefixComma=false. The reverse “HAVING … GROUP BY” loop is left untouched because it only recognises COMMA as a separator and cannot reach the no-comma branch.

• SQLASTOutputVisitor.visit(SQLSelectGroupByClause) emits a comma before a non-first SQLGroupingSetExpr sibling iff its hasPrefixComma flag is true; otherwise the historical newline-only formatting is kept. The fix lives in the base visitor so every dialect output visitor inheriting from it (ODPS, Hive, Oracle, MySQL, Spark, ClickHouse, SQLServer …) picks up the behaviour.

Adds OdpsGroupingSetsCommaTest covering ODPS exact-string roundtrip for the comma / no-comma / grouping-sets-only forms, cross-dialect smoke tests for Hive and Oracle, and AST-level invariants (default flag, clone propagation, equals/hashCode distinguishing the flag, programmatic default emit shape).

• docs(openspec): record GROUPING SETS comma fix as archived change

Adds the openspec change record for the GROUPING SETS comma fix and syncs the new scenarios into the sql-parser-core baseline spec:

openspec/specs/sql-parser-core/spec.md + Requirement “GROUP BY GROUPING SETS Separator Preservation” with six scenarios covering parse + emit roundtrip for the comma / no-comma / grouping-sets-only forms, programmatic default emit shape, AST equality, and clone propagation.

openspec/changes/archive/2026-05-12-fix-grouping-sets-comma/ proposal.md, tasks.md, verification-notes.md, .openspec.yaml and the matching ADDED-Requirements delta spec under specs/sql-parser-core/spec.md.

The change is filed directly into archive/ because it ships as part of the same commit series as the implementation; the baseline already reflects the new Requirement so there is no pending delta to apply.