Recipemd

- *Fix:* Don't parse links that partially wrap an ingredient's name as
amount links. This was never spec conformant, but the test cases
were incorrect up to spec version 2.3.5
- *Fix:* Preserve reference links and reference-style images in
description and instructions
- Switch the underlying commonmark markdown parser from
[`commonmark.py`] to [`markdown-it-py`]
- [`commonmark.py`] is deprecated and the project recommends
switching to [`markdown-it-py`]
- The new parser allows for more accurate parsing of certain
markdown constructs, enabling the fixes detailed above

[`markdown-it-py`]: https://markdown-it-py.readthedocs.io/

- *Fix:* Work around bug in [`commonmark.py`] which led to incorrect
parsing of fenced code blocks

[`commonmark.py`]: https://github.com/readthedocs/commonmark.py

- *Fix:* Make `--flatten` compatible with Python 3.9
- *Fix:* Use UTF-8 encoding when loading linked files
- *Fix:* Fix path being parsed as port under windows
- Add integrations tests for the `recipemd` CLI

- *Fix:* Show error on duplicated yields/tag instead of silently ignoring the earlier occurrence.
- *Fix:* Prevent error when scaling a recipe with a factor-less yield.
- Relax version requirement for yarl and argcomplete
- Include test cases in source distributions

- *Fix:* Fix missing package due to bundling error
- *Fix:* Incorrect heading in changelog

- *Fix:* Scaling of amounts with `None` factors
- *Fix:* Possible infinite recursion when flattening recipes
- *Fix:* Make `recipemd` compatible with Python 3.9