OTH: Update loess

[derb12]

Description

Updates loess as follows:

• allows a custom noise sigma function to perform Ruckstuhl's robust extraction of baseline signal method
  • also now guards against a noise sigma value of 0
• is more robust against windows with too few non-zero weights
  • the check for each window makes it ~10% slower than before this PR; could probably offset the slowdown by solving low polynomial orders (<2) by directly using the hat matrix in a follow up, but need to see how the numerical stability is affected. Doesn't matter a ton since delta can just be increased
• allows masking for loess as a follow up to PR ENH: Add masking support for many algorithms #83 to address issue ENH/RFC: Add masking #72
• deprecates conserve_memory and instead just always calculates the distance kernels each iteration

Type of Pull Request

• Bug Fix
• New Feature
• Miscellaneous Changes (refactor, code improvements, etc.)
• Documentation or Example Programs

Pull Request Checklist

• New code and/or documentation is valid for use with the BSD 3-clause license.
• New code is fully documented with docstrings that follow Numpy style,
  if applicable.
• New code follows PEP 8 standards as closely as possible, if applicable.
• Added/updated tests and ensured they pass locally, if applicable.
• Verified that documentation builds locally, if applicable.

[derb12]

test_polynomial.py::TestLoess::test_zero_sigma_exits_2 seems flaky; originally failed for 3.10, 3.11, and 3.13; rerunning then passes. Should either mark it as xfail or just remove it since test_polynomial.py::TestLoess::test_zero_sigma_exits does the same check; will do after merging since it's not worth another full CI run.