fixing negative power to zero

[raushanprabhakar1]

Which issue does this PR close?

• ClosesPostgreSQL compatibility: power(0.0::float8, -1.0::float8) should error, not return infinity #22272

Rationale for this change

PostgreSQL rejects power(0::float8, negative::float8) because zero raised to a negative power is undefined. DataFusion used f64::powf, which returns infinity for that case. This aligns float (and related decimal float paths) with PostgreSQL’s domain semantics.

What changes are included in this PR?

• Added float64_power_checked to error when base == 0.0 and exp < 0.0 with message zero raised to a negative power is undefined.
• Wired it into Float64/Float64power, decimal array-exponent float fallback, and compute_pow_f64_result.
• Updated math.slt expectations for pow(0, -0.5) and added coverage for power(0.0::float8, -1.0::float8).
• Added unit tests in power.rs.

Are these changes tested?

Yes. Unit tests in datafusion/functions/src/math/power.rs, and SQL logic tests in datafusion/sqllogictest/test_files/math.slt (including the regression for #22272).

Are there any user-facing changes?

Yes. Queries that previously returned positive infinity for power / pow with a zero base and negative exponent now return a runtime error (compute error) with that message, matching PostgreSQL for this case.

No public Rust API signature changes; treat as SQL/runtime behavior only unless your process labels SQL semantics as API.

[raushanprabhakar1]

Hi @Dandandan , requesting you to review. Thanks.

[raushanprabhakar1]

Hi @kumarUjjawal, Thanks for the review, extended tests as per your suggestions. Thanks

[kumarUjjawal]

Thanks @raushanprabhakar1 looks good, just one request, please don't resolve the comments yourself, I like to read through them while reviewing the newer changes.

[kumarUjjawal]

There are some discussion going on about which engine we should follow for compatibility, I think this case is pretty straightforward.

[Jefffrey]

• it seems this may have been a conscious decision in the past, see: Make log function to be in sync with PostgresSql #5259