BUG: fix segfault on deeply nested structured dtype (gh-32488) - #32622

PR summary

Fixes #32488.

Creating an array from a sufficiently deeply nested structured dtype segfaults instead of raising a catchable exception. Resolving the zerofill traverse function for such a dtype recurses once per nesting level in npy_get_zerofill_void_and_legacy_user_dtype_loop, with no termination check, so a deep enough dtype exhausts the C stack.

The fix adds a Py_EnterRecursiveCall guard via a thin wrapper around the existing function, following the same pattern used for self-containing arrays (gh-8306, gh-9077).

This fixes the crash as reported in the issue (verified against the reporter's own repro, which now raises RecursionError instead of segfaulting). It does not fully cover the artificially small (1 MiB) thread-stack reproducer also posted in the issue, that's a general limitation of counter-based recursion guards rather than something specific to this fix, same caveat noted in gh-32620.

Added a regression test in TestMonsterType following the existing test_dict_recursion/test_list_recursion style. Ran the full test_dtype.py suite (1181 passed) and the zerofill/structured/dtype subset of test_multiarray.py (195 passed) with no regressions.

First time contributor introduction

Still new to numpy, background in CPython tooling. Found this one from the "Looking for Contributor" board after that PR.

AI Disclosure

No AI tools used.