ObjectSpace._id2ref is deprecated in 4.0 — and the gate it needs was wrong already - #896

…wrong already

One more of the 27 examples from #799. `core/objectspace/_id2ref_spec.rb` goes
0 → 1; the whole file is that one example at 4.0, the eleven others being
`ruby_version_is ""..."4.0"`.

gc.c calls rb_category_warn(RB_WARN_CATEGORY_DEPRECATED, "ObjectSpace._id2ref
is deprecated"). The method still answers; it only says so.

## The find is not the deprecation, it is the gate

Writing this meant copying the nearest precedent — `iterator?`, which rbgo
already deprecates. Measuring that precedent against MRI 4.0.7 first showed it
wrong in two ways, and copying it would have spread both:

    $VERBOSE = nil; Warning[:deprecated] = true; iterator?
      rbgo  "…:13: warning: iterator? is deprecated; use block_given? instead"
      MRI   ""

MRI's deprecation_warning_enabled() wants a NON-NIL $VERBOSE *and* the
category. rbgo checked the category alone, so it spoke where MRI is silent —
and `$VERBOSE = nil` is precisely how a program asks for that silence. Measured
in all four combinations: nil/on silent, false/on warns, true/off silent,
true/on warns.

Second, with a `Warning.warn` override installed:

    module Warning; def self.warn(msg, category: nil) = $seen << [msg, category]; end
      rbgo  [..., nil]
      MRI   [..., :deprecated]

The category is not decoration: a handler that filters on it saw nil.

Both are our code, so both are fixed here rather than filed. warnDeprecated is
MRI's gate in one place — the two-condition test, the "<file>:<line>: warning: "
head, Warning.warn with category: :deprecated — and `iterator?` now goes
through it.

## Measured

rubydiff against MRI 4.0.7, three probe programs: AGREE on all of them,
including the four $VERBOSE × Warning[:deprecated] combinations and the
Warning.warn interception.

The argument check stays ahead of the warning, so `ObjectSpace._id2ref` with no
argument raises ArgumentError in silence — a warning emitted first would make a
plain mistake look like a deprecation. There is a test for it.

warnDeprecated: 100.0% covered.

## Not done here

The full internal/vm suite was NOT run locally for this change: the machine is
at a 1-minute load average of 12.5, and the one run attempted hit #895 —
TestUDPSocketLoopback hanging for 38 minutes until the suite's 40-minute
timeout. CI runs it on this pull request.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>