Fix openGauss LIMIT offset, row-count parameter marker order

[daguimu]

Problem

On ShardingSphere-Proxy against openGauss, a prepared statement that uses the MySQL-style LIMIT offset, count shortcut with parameter markers returns rows in the wrong window. For example, SELECT * FROM employees ORDER BY id LIMIT ?, ? bound with setInt(1, 2); setInt(2, 1) is expected to skip 2 rows and return 1 row, but ShardingSphere treats the first ? as row-count and the second ? as offset, so it returns 2 rows starting from offset 1.

Root Cause

OpenGaussStatementVisitor.createLimitSegmentWhenRowCountOrOffsetAbsent builds the LimitSegment for the LIMIT selectOffsetValue COMMA_ selectLimitValue grammar branch by visiting selectLimitValue before selectOffsetValue. Parameter marker indices are assigned from the visitor counter in visit order (see OpenGaussStatementVisitor.visitParameterMarker and the parameterMarkerSegments collection), not from lexical position. Visiting the row-count child first therefore gives the second ? index 0 and the first ? index 1, which is the reverse of the user-facing parameter order.

Fix

Visit selectOffsetValue first, then selectLimitValue, in createLimitSegmentWhenRowCountOrOffsetAbsent. Parameter marker indices now line up with lexical order, matching how DorisStatementVisitor.visitLimitClause already handles the identical MySQL-style shortcut.

Tests Added

Added OpenGaussStatementVisitorTest under parser/sql/engine/dialect/opengauss/src/test:

Change point	Test
Swap visit order so LIMIT ?, ? offset gets index 0 and row-count gets index 1	assertVisitLimitWithOffsetAndRowCountParameterMarkers - asserts offset.parameterIndex == 0, rowCount.parameterIndex == 1 (fails without the fix, passes with it)
Regression: literal LIMIT n, m still parses correctly	assertVisitLimitWithLiteralOffsetAndRowCount - asserts offset value 1 and row-count value 2
Regression: single-marker LIMIT ? unchanged	assertVisitLimitWithOnlyRowCountParameterMarker - asserts row-count parameter index 0 and no offset
Regression: separate OFFSET ? LIMIT ? clauses unchanged	assertVisitOffsetAndLimitWithParameterMarkers - asserts offset parameter index 0 and row-count parameter index 1

Impact

• Touches only the MySQL-style LIMIT offset, count branch in the openGauss visitor.
• Literal-valued LIMIT n, m, plain LIMIT ?, LIMIT ? OFFSET ?, and FETCH paths are unchanged.
• PostgreSQL parser is not affected (its grammar does not accept the LIMIT offset, count shortcut).

Fixes #34961