gh-150587: Factor non-negative compact int subscript guards

[KRRT7]

This PR is now a refactor/cleanup, not a bug fix.

The original version assumed there was a correctness issue around non-compact int indices reaching specialized subscript paths. That turned out to be incorrect: the existing specialization and guard structure already prevents that case from reaching _PyLong_CompactValue() unsafely.

What this PR does now is factor the "exact non-negative compact int" check into a shared guard uop and use it in the specialized subscript paths that already rely on that invariant:

• BINARY_OP_SUBSCR_LIST_INT
• BINARY_OP_SUBSCR_STR_INT
• BINARY_OP_SUBSCR_USTR_INT
• BINARY_OP_SUBSCR_TUPLE_INT
• STORE_SUBSCR_LIST_INT

The refactor keeps the specialized opcode expansions at the same uop counts as before by replacing _GUARD_TOS_INT where appropriate, rather than stacking an additional guard on top.

Tests run:

• env PYTHONPATH=build/lib.macosx-26.5-arm64-3.16 ./python.exe -m test test_opcache -m test_binary_subscr -m test_store_subscr -v
• env PYTHONPATH=build/lib.macosx-26.5-arm64-3.16 PYTHON_JIT=1 ./python.exe -m test test_opcache -m test_binary_subscr -m test_store_subscr -v

• Issue: BINARY_OP_SUBSCR_LIST_INT / STORE_SUBSCR_LIST_INT silently depend on _GUARD_TOS_INT enforcing compactness #150587

[Fidget-Spinner]

LGTM, thanks for catching this bug, just one request.

[markshannon]

This PR is marked as fixing #150587, but that issue is closed as not planned.

What is this fixing?

[KRRT7]

@markshannon that is my own issue, the original issue was for a much bigger PR and I've since then, scoped it down, will update the body / title.

[Fidget-Spinner]

Wait upon closer inspection, the current code already deopts in the subscript op if it sees a negative index. I think there's no bug here? Correct me if I'm wrong.

Also GUARD_TOS_INT checks for a compact long already, then the subscript op itself checks for the range .

[KRRT7]

You were right about the original bug claim: after rechecking the guard/specialization path, the non-compact-index issue I described is not actually reachable.

I’ve kept the branch but narrowed it into a refactor only: factor the exact non-negative compact-int check into a shared guard uop for the specialized subscript/store-subscript paths that already depend on that invariant, without increasing their uop counts.

I reran the focused test_opcache coverage for test_binary_subscr and test_store_subscr, with and without PYTHON_JIT=1.