Ufal.morphodita

------------------

--------------------------
- Fix a bug in the tokenizers, which did not end a sentence on an
empty line if the preceding line ended with whitespace. However,
only the REST server and the API was affected; the `run_*` binaries
read input a paragraph at a time, and so circumvent the problem.
- Fix warnings in Clang 15 by qualifying `std::move`.

--------------------------
- Do not assume char is signed; fixes silent errors on aarch64
when compiled from the source (e.g. in Docker on ARM macOS).
- No new binaries are provided, because none were affected.

--------------------------
- When analyzing numerals, if the form ends with a dot, keep the dot
in the lemma (previously the dot was discarded).
- Correctly handle lemma number which is not null-terminated in `generate`.
- On Windows, the file paths are now UTF-8 encoded, instead of ANSI.
This change affects the API, binary arguments, and program outputs.
- Add ARM64 macOS build.
- The Windows binaries are now compiled with VS 2019, older systems
than Windows 7 are no longer supported.
- Add billing information to the REST server.
- Add type information to the Python package.
- The Python wheels are provided for Pythons 3.6-3.11.

--------------------------
- Allow overriding path to Java library in the bindings.
- Document statistical guesser training using train_guesser binary.
- Improve the GRU tokenizer.
- Do not generate empty lemmas in guesser.
- Add explicit `GUESSER_UNSPECIFIED` guesser mode already used by tagger.
- Fix unaligned pointer access and other UndefinedBehaviourSanitizer findings.
- Fix bug in tag_filter incorrectly matching negates sets of 2+ characters.
- Add several REST server options.
- Add methods running tagset_converter during derivation_formatter.
This allows correct handling of their interactions.
- Add binary wheels of the Python package.

-------------------------
- Fix regression in `morpho::generate` caused in 1.9.1.
- Add check in `encode_dictionary` that no lemma is repeated.