Incorrect reference handling in _arraycore initialization failure paths #5271

This issue does not affect users with working NumPy installations, the affected code runs only when NumPy initialization fails. It is a defensive hardening opportunity I encountered while chasing a separate memory issue through my project's dependencies.

The affected file is Bio/Align/substitution_matrices/_arraycore.c on the current Biopython master branch.

Expected behaviour

If numpy.ndarray is missing, invalid, or unsuitable as a base type, _arraycore should abort initialization while releasing exactly the temporary references it owns.

Actual behaviour

The initializer decrements the NumPy module reference after retrieving numpy.ndarray, then decrements the same reference again when the attribute is missing or is not a type. Other early returns can retain the temporary numpy.ndarray reference.

This can corrupt or imbalance CPython reference counts during failed module initialization.

Steps to reproduce

Load _arraycore in a fresh CPython process with a stub numpy module where ndarray is either missing or not a type. Both cases exercise the incorrect cleanup paths.

I have a tested fix ready that balances the temporary references and uses PyModule_AddType for the successful ownership transfer. I can open a pull request if the maintainers would like this failure path hardened.