perf: interpreter block invocation from native iterators #382

Part of #378.

Array#each, #map and #any? over 10 elements (ns per whole call):

That is about 290 ns per element in the interpreter, versus about 36 ns under tier-2. The tier-2 path already has a cheap way to call a block from a native iterator. The interpreter does not use it.

This is the largest single interpreter gap: Rails runs 223 block calls per request, and about 40 of them come from Array#each alone.

Do:

• Profile the interpreter path.
• Port the tier-2 native-iterator → block call path (or its essence: frame reuse across iterations, no per-yield re-dispatch) to the interpreter, for Array/Hash/Range/Integer iterators first.

Done when:

• Interpreter each(10) costs ≤ 2× CRuby.
• No diff_cruby regressions, especially break/next/redo/return-from-block and ensure fixtures.