Agilkia

Added
- PrefixTraceExtractor supports more flexible ways of extracting ML features from traces.
- Added eShop example: trace file and some scripts
- Updated dependency versions to better support Python 3.8 and higher.
- Added developers_agilkia.yml so it is easy to create an Agilkia development environment.
- Added CHANGELOG.md

Changed
- Restructured the main docs page so it is more useful on ReadTheDocs.io.
- Improved the docs of TracePrefixExtractor to show new features.
- Added --cluster option to read_scanette_csv.py script, which does bag-of-words MeanShift clustering.

Removed
- Nothing.

Added
- Support for hierarchical clustering in output JSON files. See TraceSet.set_clusters etc.
- Added example of hierarchical clustering in examples/scanner/cluster_hierarchical.py.
- Add examples/scanner helper scripts for converting to/from Scanette *.csv files.
- Add examples/scanner scripts to illustrate clustering and choosing a given number of tests.
- Added Event.status_float for non-integer status values, with NaN for missing/strange values.

Changed
- Improved the tests for reading various older versions of JSON files.
- Improve API of TraceSet.with_traces_grouped_by to take a lambda function for the key,
and to throw exception on missing keys by default.
- Cluster labels are now stored in the JSON file, with an optional scipy linkage array for hierarchical clusters.

Changed
- Bug fix release to correct dependencies. Removed fastcluster package as it clashed with some numpy versions.

Added
- Added examples/scanner for the SuperMarket scanner example (scanette in French).
- Nicer printing of Event objects.
- Added TracePrefixExtractor (an sklearn feature extractor for trace prefixes) to support machine learning.
- Added SmartSequenceGenerator that uses an ML model to generate sequences of actions.
- Added SmartSequenceGenerator.generate_all_traces algorithm, to generate all high-probability traces.

Changed
- Allow SmartSequenceGenerator to take URLs, and add a way of executing generated tests.
- Fixed errors in decode_outputs, to handle zeep objects properly.
- Change TraceSet.clusters to be private (._clusters) so it is not saved.
This is for consistency, since we cannot easily save _cluster_data and both are needed for visualize_clusters.
The idea is that clustering is transient, so needs to be redone after save/load.
- Improved the default copying of meta_data when a TraceSet is created with traces all from one parent.

Added
- Added support for readthedocs so that documentation is built automatically for each version.

Changed
- Added consistent meta_data field to all three levels: Event, Trace, TraceSet.
- Made clustering and visualisation a bit more flexible.

Added
- Add more meta-data to the flit pyproject.toml file.

Changed
- Delay import of arff until user calls to_arff, because liac-arff is pip-only and conda-build does not seem to handle that.
- Documentation in Sphinx format, to go onto readthedocs.org.