ENH: template-ify lambertw with overloads for double, float, and complex<float>

[kandersolar]

This PR adds overloads for lambertw for double, float, and complex<float> inputs. The implementation follows @steppi's suggested approach in this forum thread: https://discuss.scientific-python.org/t/scipy-special-faster-real-only-versions-of-special-functions/2271/4

Please note that this is my first time dealing with modern C++. Please review the code assuming it was written by someone that figured it out as they went :)

I can say at least that it compiles and passes pixi run tests, which I take to mean that I did not break the existing complex<double> path. However, I'll need some guidance with creating tests for the new dtypes. I see there are parquet files with test values, but I'm lost beyond that.

[steppi]

Thanks @kandersolar. For testing, the parquet files contain test cases from the entire scipy.special test suite and the tests and files were generated automatically. Check out tests in xsf/tests/xsf_tests for examples of test code which was added manually. This should be fairlystraightforward to review; I will try to take a look within the next week or two.

[steppi]

The windows failures are known and unrelated. You don't have to worry about the formatting issues for now.