Implement native support for percent_rank window function

[weimingdiit]

Describe
Spark percent_rank() is not supported in Auron's native window execution path, so queries using it fall back instead of being executed natively.

Describe the solution you'd like
Add native support for Spark percent_rank() in Auron's window execution.

The implementation should preserve Spark semantics for:

• partitioning
• ordering
• peer row handling
• return type and edge-case behavior