Stumpy

--------------------------
* bugfixes
- None
* features
- None
* tasks
- Bumped minimum NumPy version to use `np.array_equal`
* documentation
- None

--------------------------
* bugfixes
- None
* features
- Added maximum window size check
- Added window size checking to preprocessing
* tasks
- Replaced array comparison in `core.are_arrays_equal` with `np.array_equal`
- Added Github Actions and badge
- Improved stability for integration with other packages
* documentation
- Fixed typo
- Replaced NABDConf motivation video with PyData Global video

--------------------------
* bugfixes
- Fixed inconsistent `mstump`/`mstumped` output to match description in published work
* features
- Added `subspace` function for compute multi-dimensional matrix profile subspace
- Added `include` and `discords` handling to `subspace`
- Added `mdl` building blocks
- Added `ostinato` Dask distributed, GPU, and `aamp` variants
- Added fast rolling min/max functions
* tasks
- Updated Azure Pipelines CI coverage tests to use latest images
- Added `mstump` tutorial to Binder
- Fixed bad reference to `aamp_gpu_ostinato`
- Fixed `aamp_ostinato` unit test coverage
- Converted `conda.sh` to `bash`
- Moved all private functions out of `__init__.py`
- Added commands to remove `build/`, `dist/`, and `stumpy.egg/` after PyPI upload
- Added twine to environment setup script
- Fixed incorrect date in `CHANGELOG.md`
* documentation
- Added `mstump` tutorial to RTD
- Updated various function docstrings
- Added Github Discussions to README and RTD
- Updated API list on RTD

--------------------------
* bugfixes
- Fixed incorrect cancelling of Dask data that was being scattered with `hash=False`
- Fixed floating point imprecision in computing distances with `mass_absolute`
* features
- Added `ostinato` function for computing consensus motifs
- Added new approach for retrieving the most central consensus motif from `ostinato`
- Added `mpdist` function for computing the MPdist distance measure
- Added `mpdisted` function for computing the MPdist distance measure
- Added `gpu_mpdist` function for computing the MPdist distance measure
- Added `aampdist` function for computing the MPdist distance measure
- Added `aampdisted` function for computing the MPdist distance measure
- Added `gpu_aampdist` function for computing the MPdist distance measure
- Changed `np.convolve` to the faster `scipy.signal.convolve` for FFT
- Added matrix profile subspace for multi-dimensional motif discovery
- Replaced existing rolling functions with fast Welford nanstd and nanvar functions
* tasks
- Updated Azure Pipelines CI to use latest images
- Fixed tutorial typos
- Added ostinato paper to README References
- Added `mamba` to speed up Python environment installation
- Added `ostinato` tutorial
- Added a series of new unit tests (maintained at 100% test coverage)
- Updated all tutorials to use Zenodo links for data retrieval
- Removed tutorial plotting function that set default plotting conditions
- Added AB-join tutorial
- Replaced rolling window isfinite with a much faster function
- Updated Github PR template to use conda-forge channel
- Updated Welford corrected-sum-of-squares derivation
- Updated Binder environment to use STUMPY release in Binder postBuild
* documentation
- Fixed GPU function signatures that were being displayed on RTD
- Fixed incorrect docstring indentation
- Added STUMPY docs and Github code repo to Resources section of tutorials
- Added default values to docstrings

--------------------------
* bugfixes
- Fixed AB-join so that it now matches the published definition (previously, BA-join)
* features
- Added nan/inf support to FLOSS/FLUSS
* tasks
- Removed Pandas series in GPU tests to improve unit test speed in CI
- Identify operating system prior to installing cuda toolkit
- Changed `left`/`right` keywords in all unit tests to `ref`/`comp`
- Split up unit tests and coverage testing
- Updated `displayNames` in Azure Pipelines
- `test.sh` now accepts `unit`, `custom`, and `coverage` keywords
- Fixed typos
- Added pattern searching (MASS) tutorial
- Added `Contribute` notebook to RTD table of contents for first time contributors
- Refactored `_compute_diagonal` for speed improvements
- Replaced `np.roll` with slice indexing in `stumpy.floss`
- Refactored and improved `aampi` and `stumpi` update performance
- Added `lxml` to environment.yml
* documentation
- Added `plt.show()` to code figures in tutorials
- Updated `stumpi` tutorial with `egress=False`

--------------------------
* bugfixes
- Fixed warning and check when time series has inappropriate dtype
- Fixed scenarios where identical subsequences produce non-zero distances
* features
- For interactive data science work, matrix profile calcs are 10-15x faster
- Added `aamp` with non-normalized Euclidean distance (i.e., no z-normalization)
- Added `aamped`
- Added `aampi`
- Added `gpu_aamp`
- Added egress for `stumpi` and egress is now the default behavior
- Added a `mass_absolute` function for non-normalized distance calculation with FFT convolution
- Added diagonal pre-processing function to `core.py`
- Added centered-sum-of-products and Pearson correlation in place of sliding dot products
- Added left and right matrix profile indices to `scrump` and converted to Pearson correlation
* tasks
- Removed Pandas series in GPU tests to improve unit test speed in CI
- Updated to latest version of black for better formatting
- Refactored redundant test section
- Added unit test for inappropriate dtype inputs
- Corrected absolute stumpy import to be relative import
- Replaced `._illegal` attribute with a more obvious `._T_isfinite` attribute
- Moved common diagonal functions to `core.py`
- Replaced `order` variable with the more obvious `diag` name
- Added environment.yml for easier installation of dependencies
- Removed random print statement in code
- Moved STUMPY thresholds to global parameters in `config.py`
- Refactored left/right matrix profile indices
- Refactored NaN checking
- Check for Linux OS and add TBB dynamically especially for CI
* documentation
- Added `aamp` reference to README
- Update docstrings to be less verbose for API documentation
- Fixed some typos
- Replaced `sep="\s+"` with `sep="\\s+"` in tutorials
- Added notes and derivations for Pearson correlation and centered-sum-of-products
- Renamed tutorials with underscores for consistency
- Added all `aamp`-like functions to API reference
- Replaced MS Word docs with LaTeX notebooks