Light-curve-python

Added

- `check: bool = True` keyword argument for `__call__` and `many` methods of feature classes. It coulb be used to check
if input arrays are valid

Changed

- `gsl` is a default Cargo feature now, which means that GSL must be installed to build this package by standard Python
tools like `pip install`
- `light-curve-feature` 0.3 -> 0.4.1 transition brings MCMC improvements, changing feature names of `BazinFit` and
significant changes of `VillarFit` feature set

Removed

- `antifeatures` submodule of the Rust implementation

Fixed

- Fix implementation of `OtsuSplit`, see [issue 150](https://github.com/light-curve/light-curve/issues/150)

Added

- Pure-Python implemented `OtsuSplit` feature extractor (experimental)
- Python snippets in `README.md` are tested, this requires `pytest-markdown`
- A lot of new unit-tests for the pure-Python implementation
- New benchmarks to compare pure-Python and Rust implementations performance
- Publish Linux packages for `aarch64` and `ppc64le`

Changed

- The Python package is renamed to `light-curve`, `light-curve-python` still exists as an alias
- Pure Python implementation of the most of the features are added, now Rust-implemented features live
in `light_curve_ext` sub-package, while the Python implementation is in `light_curve_py`. Python-implemented feature
extractors have an experimental status
- Now `dataclasses` (for Python 3.6 only) and `scipy` are required, they are used by the pure-Python implementation
- `maturin` version 0.12

Fixed

- The Rust implemented classes `.__module__` was `buitins`, now it is `light_curve.light_curve_ext`

Changed

- Rust edition 2021
- Minimum supported Rust version is 1.56
- Maturin version 0.11

Fixed

- An exception shouldn't be raised for the case of small time-series length and non-`None` `fill_value`

Added

- Support `dtype=np.float32` for feature extractors
- `_FeatureEvaluator.many(lcs)` method for parallel execution