From TF meeting 2022-11-01 #343, agreed to update the document with:

• adding corresponding character classes for non-wide width ones over defined classes, A/B/BA - adding Ap or something
• expand existing table for omitting spacing, using newly added character classes, with copying conditions where appopriate

key aspect is disambiguation of Roman chars from Roman (aka proportional or half-width) punctuation, so the latter can be set flush while the former gets J-Roman aki. And, tsume can happen on the full-width punctuation but none on Roman punctuation.

I cannot read Japanese, so I don’t know if this has been addressed.

But recently I was working on something like this (in Lua) and discovered some CJK punctuation can be full-width or half-width, depending on the typeface/language. The specific problem characters I was running into were U+3008 and U+3009, which I understand would be classed as cl-01.

key aspect is disambiguation of Roman chars from Roman (aka proportional or half-width) punctuation,

As far as following current definition, all of A/B/AB has "so-called normal full-width" condition, so I believe such cases are taken up in that table.
For proportional Japanese fonts, user agents shall take care of how much spacing is in glyph, and nothing could be taken account in collapsing rules. These are the same as non-proportional ones, but are just a matter on width of spacing included like 0 to 0.5 em for proportional against exactly 0.5em for non-proportional.

But recently I was working on something like this (in Lua) and discovered some CJK punctuation can be full-width or half-width, depending on the typeface/language.

These might be 'proportional' fonts. Check something like 'MS P Gothic' (proportional for 'MS Gothic').