Breaking changes
* `typing_extensions>=4.5` is now a required dependency on `python<3.13`
* Caching is now done automatically by default, i.f.f. `r + sum(trim) <= 24`, and the trim is integral. For the previous default behavior, pass `cache=False`.
* The `rowvar` parameter for the L-comoment sample estimators is now automatically determined by default. Previously, the default was `rowvar=True`.
* The deprecated `lmo.distributions.l_rv_nonparametric` has been removed (as planned). Use `lmo.distributions.l_poly` instead.
Fixes
* Fixed several caching issues in `lmo.l_weights`. The consequence is that, all sample estimators will now be faster, more precise, and can handle larger orders.
* The type annotations and the (undocumented) `lmo.typing` module, have been fully reworked. This fixes many typing issues. Runtime behavior shouldn't be affected (in theory).
* Several documentation inconsistencies have been corrected.
Added
* `l_kurt` and `l_cokurt` aliases for L-(co)kurtosis.
* Type-tests to validate correct `overload` and generic behavior of relevant `lmo.l_*` estimators.
* More & better hypothesis tests, specifically for `lmo.l_weights`.
**Full Changelog**: https://github.com/jorenham/Lmo/compare/v0.13.2...v0.14.0