perf: ensure same hidden class for OperatorSubscriber

[cartant]

Description:

See the comments within the change. This PR should help realise one of the advantages of replacing operator-specific subscribers with a general-purpose subscriber - reducing/eliminating megamorphic property accesses within Subscriber.

I've not done any micro optimization/analysis, yet. I'll look at doing that after the version 7 release. However, I think that what's in this PR makes sense and shouldn't be controversial. 🤞

Related issue (if exists): None

[cartant]

I don't understand this CI failure. It seems to have something to do with side effects? I ran test:side-effects:update, but it seemed to generate a surprising number of changes, so ... 🤷‍♂️

[benlesh]

So the CI errors are from the side-effects checker. However, I think if you implement the change I'm suggesting (which is to use super._next, super._error, etc), then you won't have those declarations at the top level, and the side-effects checker will be happy again.

[cartant]

So the CI errors are from the side-effects checker.

Yeah, I can see that it's related to that: the checker has no way of knowing (with absolute certainty) that the property accesses are not calls to getters. (It might not know that these aren't side effects, but we do.) I was mostly concerned about the diff after running the update script. It seemed to be unnecessarily large. I'll have another look at it after #5898 is merged/addressed, as that problem is now breaking the build.

[benlesh]

Sorry, this one almost fell off my radar, @cartant. Thanks for the work!