According to the MDN article:

https://developer.mozilla.org/en-US/docs/Web/API/Document/querySelector
> CSS Pseudo-elements will never return any elements,
> as specified in the Selectors API

which refers to (older, level1) Selectors API:
https://www.w3.org/TR/selectors-api/#grammar
> Authors are advised that while the use of pseudo-elements in selectors
> is permitted, they will not match any elements in the document, and thus
> would not result in any elements being returned. 

Rune introduced pre-filter of pseudo elements for querySelector in
https://codereview.chromium.org/1602833002

I cannot find the original note which was in older spec in the latest
selectors level4 spec, though:
https://drafts.csswg.org/selectors-4/

Anyway, to allow returning what is expected here, we have several options
- explicitly exclude ::slotted() in filtering for querySelector
- change ::slotted() to something other than pseudo-element (e.g. combinator)
- etc.

Looks like querySelector() API is split out of selectors-api spec, and
the selectors-4 spec is purely for CSS selectors, and querySelectorAPI()
is now part of DOM spec: https://dom.spec.whatwg.org/
But I cannot find any notes on matching pseudo-elements in querySelector().

So if excluding any pseudo-elements in querySelector is not a *requirement*,
then we can return matched elements for ::slotted(), just like what we do
for '>>>' (shadow-piercing combinator) use in querySelector, without changing
any spec.

Maybe ::part() should match as well, once CSS shadow parts (formerly called
custom pseudo element) is implemented.
https://tabatkins.github.io/specs/css-shadow-parts/

kochi@, could you work on this?

Upstream spec issue filed at:
https://github.com/whatwg/dom/issues/463

Once upstream issue is decided whether we allow returning elements which
live in different tree than the context object for querySelector() et al.,
I'll work on matching the spec.

The upstream issue is closed and this is working as the spec says.