BUG: fix two error handling mistakes in stringdtype replace loop - #32355

PR summary

Two error paths in the StringDType replace implementation had bugs. One one is user-visible via a public API:

In [1]: arr = np.array(["aaa"], dtype="T")

In [2]: np.strings.replace(arr, "X", count=-12)
Out[2]: array(['aaa'], dtype=StringDType())

This should be array(['XXX'], dtype=StringDType()), for any negative count. Right now it only happens if count == -1. This follows python strings:

In [3]: "aaa".replace('a', 'X', count=-12)
Out[3]: 'XXX'

The other one requires passing out to the private _replace ufunc directly to have a user-visible impact. Still, seems worth adding the goto fail that's missing.

AI Disclosure

An AI model spotted both issues.