Add kernel weighting functions

[thisisamirv]

Summary

This PR adds a new module kernel_weights providing common statistical kernel functions used in local regression and kernel density estimation.

Details

• Implements five kernels:
  • tricube
  • epanechnikov
  • gaussian
  • triangular
  • quartic (biweight)
• Adds corresponding types (Tricube, Gaussian, Epanechnikov, Triangular, Quartic) implementing a new KernelFn trait.
• Allows fn(f64) -> f64 to be used directly as KernelFn.
• Includes integration and behavioral tests ensuring normalization and symmetry.
• Fully documented with Rustdoc examples and #[must_use] attributes.

Motivation

Kernel weighting functions are widely used in nonparametric statistics (LOESS, KDE, SVMs).
Adding them directly to ndarray-stats provides a reusable foundation for local regression and density estimation tools without external dependencies.

Checklist

• Tests added and passing (cargo test --test kernel_weights)
• cargo fmt and cargo clippy --all-targets -- -D warnings clean

[nilgoyette]

Why are you creating a constant version of those functions? Why can't the user use Tricube instead of TRICUBE?

[thisisamirv]

Thanks for the feedback!
My goal with the constants was mainly to provide a small ergonomic convenience, letting users treat kernels as predeclared value singletons (e.g., smooth(TRICUBE, data)) instead of having to instantiate the unit structs directly. Although, I’m totally fine with removing them if the preference is to keep the API minimal and rely on the unit structs directly. I’ll update the PR accordingly if that’s the desired direction.

[nilgoyette]

I'm not a maintainer here, I was just curious. Don't change anything for me :)