fix: use _comp_compgen with API changes

[akinomyoga]

These are the cases where I modified interfaces of existing functions.

Some involve internal generators, but I haven't applied the possible suggested internal generator syntax _comp_compgen -i CMD NAME in #962 (comment). If the syntax would be finally employed, I'll update them in a future separate PR.

[scop]

Left some comments, not a through review yet, but these can already be looked into.

One conflict to resolve, too.

[scop]

An unrelated note, but now that we're sprinkling local ret quite a bit and more to come, perhaps adding it to the standard local cur prev words cword was_split comp_args everywhere would not be a bad idea.

[akinomyoga]

This is just my preference, but I'd like to keep the variable scope small generally. Maybe this results in possibly multiple declarations of ret in a function, but I kind of accepted and got used to it because it doesn't cause a problem in Bash.

Including it in the standard variable list might be useful to prevent the variable leak in the case where one forgets to declare ret in calling a function that sets ret. But in another way of thinking, that hides the bug that one forgot to declare ret. I cannot decide which one is better for this point.

[scop]

LGTM, good to merge once tests start to pass 👍