fix ^ evaluates as bitwise XOR instead of exponentiation

[xiedeyantu]

Which issue does this PR close?

• Closes PostgreSQL compatibility: ^ evaluates as bitwise XOR instead of exponentiation #22252.

Rationale for this change

In PostgreSQL, the ^ operator represents exponentiation, but DataFusion was interpreting it as bitwise XOR. This caused PostgreSQL-dialect queries such as SELECT 2 ^ 3; to return 1 instead of 8.

This change aligns DataFusion's PostgreSQL SQL semantics with PostgreSQL for this operator while preserving existing non-PostgreSQL behavior.

What changes are included in this PR?

• Added PostgreSQL-specific handling for BinaryOperator::PGExp during SQL expression lowering.
• Lowered PostgreSQL ^ expressions to the built-in power(left, right) scalar function instead of treating them as bitwise XOR.
• Kept existing generic-dialect ^ behavior unchanged.
• Kept PostgreSQL bitwise XOR behavior unchanged via #.
• Added a sqllogictest regression covering SELECT 2 ^ 3; under the PostgreSQL parser dialect.

Are these changes tested?

Yes.

Added a regression test in scalar.slt to verify that PostgreSQL-dialect ^ is evaluated as exponentiation.

Validated with:

cargo test -p datafusion-sqllogictest --test sqllogictests scalar

Are there any user-facing changes?

Yes.

For the PostgreSQL SQL parser dialect, ^ now behaves as exponentiation instead of bitwise XOR, matching PostgreSQL semantics. Generic-dialect behavior is unchanged, and PostgreSQL bitwise XOR remains available through #.

[Dandandan]

Shouldn't it be handled somewhere else?

Do we neef to update documentation / tests, also covering bitwise XOR?

[Dandandan]

Ah I see it correctly uses # in the tests, so the previous behavior was probably an oversight.
I think we should move the change to self.parse_sql_binary_op(&op) though?

[xiedeyantu]

I've moved the method to binary_op.rs because BinaryOperator::PGExp doesn't have an operator chain like "#", so I used this approach to solve the problem. I also tried implementing "^" in a similar way to "#", but it required a lot of modifications and ultimately still relies on the "power" function. Would you accept this modification?