Optimize serialized size for PRISM_SERIALIZE_ONLY_SEMANTICS_FIELDS=1

[eregon]

From #2924 (comment), average went from 0.799 before #2924 to 0.950 after, with this PR it goes back to 0.800.

As an aside this makes the flags field public again, which is necessary as TruffleRuby uses it like that (the error can be seen at https://github.com/oracle/truffleruby/actions/workflows/prism.yml).

[kddnewton]

I'm okay with this, but I was under the impression you wanted the flags for the static literals maybe? Do you want to just exclude node id?

[eregon]

I was under the impression you wanted the flags for the static literals maybe?

I thought initially that would be nice mostly for the Ruby API (#1531) as a module/interface which guarantees a value method (which some of these nodes already have).

I suppose it could be useful for Java compilation to know if a node is a static literal for which the value can be potentially cached in the IR, but in practice that still depends on implementation and representation details so I think the value is fairly limited and at least currently not worth the serialized size overhead.

[eregon]

I did a quick search and for instance the only place we do instanceof Nodes.IntegerNode in TruffleRuby is for https://github.com/oracle/truffleruby/blob/3cd422433deebe3fa664f8c4540811c42ca02e93/src/main/java/org/truffleruby/parser/YARPTranslator.java#L3769-L3785 which doesn't map so nicely to the static literal flag (it covers more than that like ivar reads).

[eregon]

And for completeness, regarding the newline flag, I looked a bit into it today (previous comment about it in #2924 (comment)) but there seems to be no value to it currently in the Java API since it wouldn't really save any work from MarkNewlinesVisitor. (It could be helpful when using something else than MarkNewlinesVisitor to assign the real newline flags though I suppose, but Loader currently always uses MarkNewlinesVisitor).