• Closes range_iterator.__reduce__ uses original range + index state instead of rebased range + None #8377
• This PR follows our AI policy

Summary

range_iterator.__reduce__ (and longrange_iterator.__reduce__) returned the original range plus the current index as the pickle state, whereas CPython returns the remaining range rebased to the current position with a None state. The round-trip result was already correct on both (RustPython restored the index via __setstate__), so this is a representational / cross-compatibility divergence, not data loss.

it = iter(range(10)); next(it); next(it); next(it)
it.__reduce__()
# before: (<built-in function iter>, (range(0, 10),), 3)
# after:  (<built-in function iter>, (range(3, 10),), None)   # matches CPython

Cause

range_iter_reduce embedded the full original range and passed index as the third tuple element. Both PyRangeIterator::__reduce__ and PyLongRangeIterator::__reduce__ go through it.

Fix

Rebase the range start by index * step and emit None for the state. The index is clamped to the length first, because RustPython's iterator increments the index unconditionally, so it can run past the length after exhaustion (unlike CPython, which stops at the length). __setstate__ is left intact, so pickles that carry an integer state still load.

Test

Verified against CPython 3.14.6 — __reduce__ output now matches for mid-iteration, fresh, reversed, exhausted (range(2, 2)), step > 1, empty, single-element, negative-step, longrange, and __setstate__-advanced iterators. pickle.loads(pickle.dumps(it)) round-trips correctly in all cases.

• test_range: SUCCESS (29 run, 2 skipped — the two skips are a separate, pre-existing __setstate__ crash, unrelated to this change).
• cargo build / cargo clippy -p rustpython-vm / cargo fmt --check: clean.

No @expectedFailure marker flips here: CPython's own test_range has no test asserting the __reduce__ shape, so this is a representational fix, covered for regressions by the existing pickle round-trip tests.