Improve modeling token nodes

[Matejkob]

With @ahoppen we identified potential confusion and overcomplication in code relating to modeling token nodes. Specifically, we're currently trying to collapse multiple token kinds into a single one without indicating this in our variable names.

The only non-trivial place that needs to be updated is Child.type in SyntaxBuildableChild and, IMHO, the confusion stems from the fact that we try to collapse multiple token kinds into a single one without giving any indication of doing so in the variable names.

I think the way to go about fixing this is to change the associated value of SyntaxOrTokenNodeKind.token to [TokenChoice]. Then we won’t need to call Child.tokenKind in Child.type anymore and could shift the computation that’s currently done in this property down to uses where there’s a lot more context. Potentially, the computation might also just become a lot simpler once we see that context.

Originally posted by @ahoppen in #2006 (comment)

In relation to issue #1978, changes were made in PR #2006 to the Child.tokenKind implementation. However, this solution still appears to be incorrect and needs further attention, which this issue aims to address.

[ahoppen]

Tracked in Apple’s issue tracker as rdar://113567645