Fix usage of rb_struct_initialize() to pass an Array of members values and not a Hash and add TruffleRuby in CI

[eregon]

• rb_struct_initialize() does not accept a Hash, and it's very brittle to pass [{...}] and to rely on that C function using rb_keyword_given_p(). It basically worked accidentally, by having **members in the caller of the caller. Such logic when Struct#initialize is defined in Ruby (as in TruffleRuby) is basically impossible to implement, because it's incorrectly treating positional arguments as keyword arguments.
• rb_struct_initialize() is used in CRuby to set members of Data instances in marshal.c (there is no rb_data_initialize() yet). There, the code passes an Array of members values for Data (and for Struct which are not keyword_init: true): https://github.com/ruby/ruby/blob/48c7f349f68846e10d60ae77ad299a38ee014479/marshal.c#L2150-L2176 So we should do the same in psych.
• Rename to init_data since it's only used for Data.
• See Data object encoding #692 (comment).

[eregon]

I added TruffleRuby in CI in this PR since it just needs a tiny extra fix to be fully green.
This will help to prevent issues like truffleruby/truffleruby#3997 (or make them much faster to notice & fix)

[eregon]

We could also use members.values but then we should check if the data_class.members == members.keys, and raise an exception if not. If so, which exception class should I use then, do we have one for "incompatible change in loaded classes"?
Otherwise if e.g. one changes the order of fields from Foo = Data.define(:a, :b) to Data.define(:b, :a) then it would load the instance incorrectly in a rather surprising manner.

If there are extra members in Data.define, the current approach sets them to nil, like before:

irb(main):002> Psych::VERSION
=> "5.2.6"
irb(main):011> Foo = Data.define(:a,:b)
=> Foo
irb(main):012> d=Psych.dump Foo.new(1,2)
=> "--- !ruby/data:Foo\na: 1\nb: 2\n"
irb(main):013> puts d
--- !ruby/data:Foo
a: 1
b: 2
=> nil
irb(main):014> Foo = Data.define(:a,:b, :c)
(irb):14: warning: already initialized constant Foo
(irb):11: warning: previous definition of Foo was here
=> Foo
irb(main):015> Psych.unsafe_load d
=> #<data Foo a=1, b=2, c=nil>

This is rb_struct_initialize()-specific behavior BTW, Data#initialize doesn't allow that:

> Data.define(:a,:b).new(a: 1)
(irb):3:in 'Data#initialize': missing keyword: :b (ArgumentError)

Checking members would raise, unless we get fancy and only compare the first N members where N is members.size.

Basically this seems a good approach and it's compatible with what was done before.