Sorry @muzimuzhi, took a bit longer, but now I have shed some light again into this issue.

@hansonchar is right that there are missing occurrences of needed libraries. All of that started in issue #640. But at some point a question arose which (unfortunately) never was answered. So the work didn't continue.

When I remember correctly, the most important point why the work didn't continue was, that the extract.lua isn't able to also extract the "example"/"examples" in the Lua documentation (like the ones modified in the PR). This would help a lot, because it would enable (semi-)automatic testing. Unfortunately I don't have any clue about Lua programming :(

There are much more examples lacking libraries. See e.g. PR #883 which isn't merged yet (reason unknown anymore). And there is also Mo-Gul@af40f01 which could be cherry-picked. And it is very likely that there will be missing even more. If only extract.lua would handle these examples, too ... ;)

Please also note that the "extractor scripts" (Bash version and PowerShell version) need to be adapted to work after the "flattening of the file structure" ...

This PR is tangentially related to #1339

I think we can make small, incremental changes iteratively, instead of blocking these changes (that absolutely improve the pgfmanual) on the basis of the lack of a perfect solution.

What do you think?

In any case, I'll try to take a look at the extractor script problems later on. One step/issue at a time.

As an interesting trivia, I think I've found where the name Kellermann originated from. Till Tantau's implementation of the Tree Layouts follows the paper "A. Brüggemann-Klein, D. Wood, Drawing trees nicely with TEX, Electronic Publishing, 2(2), 101–115, 1989.", and "Kellermann" with double n's are given in Figure 10-12 in that paper, where "Kellermann" has a total occurrence of five times.