gh-129068: Make range iterators thread-safe

[colesbury]

Now that we specialize range iteration in the interpreter, there's not a significant performance cost to making the iteration thread-safe.

• Issue: Race in concurrent iteration over range iterators #129068

[colesbury]

https://github.com/facebookexperimental/free-threading-benchmarking/tree/main/results/bm-20251216-3.15.0a3+-afd3eff-CLANG,NOGIL shows 1.001x faster, which I think is just noise.

I also ran https://github.com/colesbury/fastbench/ locally:

• No measurable difference in performance
• rangeiter_next is hit 40,000 times.
• Some of those are pre-specialization. Most of them are fromlist.extend(range(...)).
• At ~10 ns overhead per critical section and 13.1 seconds benchmark run time, that's about 0.003% extra overhead

FWIW, I used Codex to convert rangeobject.c to Argument Clinic and it worked pretty well.

[kumaraditya303]

LGTM