Adding LilyPond output images

Note to self on how this should be implemented:

• It should automatically download LilyPond binaries in the doctree cache directory if not already there, to have the LilyPond version in conf.py the single source of version configuration,
• An open question is what nodes to output. https://groups.google.com/g/sphinx-users/c/3jYD9k3WxuQ
• It should be parallel. Unlike lilypond-book, it should not split into jobs from the start, but instead send a new file to each job once it has finished its current file. That way, it won't be suboptimal due to possible random differences in the total time needed for the files of each job.
• The build should be separate from the main HTML build, i.e. with a different builder, albeit with the same output dir. Alternatively, there could be a lilypond_skip_generation config option to skip producing/updating the images. It should remain possible to have a fast build without generating the images.
• The images should persist between builds. Preferably, the cache would be shared across languages, probably by reusing the same doctree dir.