FWIW, to conclude this post, and maybe of use for others, I’ve RTFM, here is the conclusion;)
Actually, there is a fantastic FAQ by @eelik that explains how solder masks work in KiCad:
The crucial piece I was missing is right at the top:
In KiCad (and other EDA) the graphical layer which represents solder mask is negative. Where there is graphics in this layer, the physical solder mask isn’t applied and there will be a hole in the solder mask.
Negative! Right. And this is specific to the mask layers, and only those.
Then, the article linked by @JohnSG clarified other confusions of mine, including this basic one:
A PCB surface finish is a thin layer applied to the exposed copper surfaces (exposed traces, pads, ground plains and holes) of a printed circuit board.
All exposed elements. Right. And “exposed” is controlled by solder mask layers (but that cannot control holes/vias, those are always exposed without special manufacturing)
Electroless Nickel Immersion Gold (ENIG) is a popular PCB surface finish that consists of two layers: an electroless nickel layer and a thin layer of immersion gold.
and
Nickel, in its pure form, is magnetic. However, in the context of ENIG (Electroless Nickel Immersion Gold) as a PCB surface finish, the nickel layer used is typically a specific type of nickel known as electroless nickel, which is non-magnetic. The reason it’s non-magnetic is that it lacks the ferromagnetic properties found in pure nickel.
So, while pure nickel is magnetic, the electroless nickel used in the ENIG process, particularly when combined with the gold immersion, does not exhibit significant magnetic properties. This makes ENIG suitable for RF applications and PCB antenna traces where non-magnetic materials are preferred.
Source: ChatGPT