GeezerVison

Since the Pratt rewrite (5.4), ^ has the same precedence as *, /, % and DIV, so 2 * 3 ^ 2 parses as (2 * 3) ^ 2. Both readings of ^ disagree with that:

5.3 got this right: BitwiseXor was the operand of MultiplicativeExpression. A PostgreSQL query like SELECT pi() * r ^ 2 now gives visitors a BitwiseXor at the top of the tree with a Multiplication inside it. The SQL text still round-trips, because the deparser prints the tree back in the same token order.

The change moves OP_CARET in prattArithRest to the level of ** (prec 7). That is above * / % DIV and still left-associative. ** already sits at that level, so in DuckDB ^ and ** (both power) now share a level, as they do in DuckDB itself. #2795 had pointed out that the two differed. No productions or lookaheads change. Only one precedence constant moves.

One thing I'm unsure about: in T-SQL, ^ sits at the + - level, below *. So with Dialect.SQLSERVER, 2 * 3 ^ 2 was right before this change and is wrong after it. 1 + 2 ^ 3 was already wrong there and stays wrong. Release 5.3 behaved like this PR. If you'd like, I can add a Dialect.SQLSERVER branch that keeps ^ at prec 5.

Tests: two new cases in ExpressionPrecedenceTest, covering *, / and % against ^, ^ on the left, left associativity, and a parse/deparse round trip.

What was run:

• ./gradlew test --tests ExpressionPrecedenceTest without the grammar change: testCaretBindsTighterThanMultiplication fails (expected: <Multiplication> but was: <BitwiseXor>). With the change, it passes.
• ./gradlew check (JDK 21, with a JDK 17 toolchain): BUILD SUCCESSFUL. That covers 10,747 tests with 0 failures, plus checkstyle, PMD, spotless, spotbugs and jacoco coverage verification.
• A differential check against PostgreSQL 17. About 5,700 random integer and boolean expressions using + - * / % ^ & | # << >>, comparisons, BETWEEN, IN, IS NULL, NOT, CASE and ::int were parsed with JSqlParser. Each tree was re-printed fully parenthesised, and PostgreSQL evaluated both forms. With this change, every remaining mismatch involves & | # << >> against + -, which is a separate issue. None is caused by ^.

Not run: gradle jmh. The change only renumbers a precedence constant, so I didn't benchmark it. checkGrammarAmbiguity: no productions changed.