BUG: fix swallowed cast error in fancy indexing assignment (#31975) - #32115

Backport of #31975.

PR summary

Fixes #31974.

arr[idx] = vals can swallow a cast error when the values are buffered and the cast fails while the iterator refills a buffer chunk (array bigger than one 8192-element chunk). the buffered iternext returns 0 on error which looks exactly like a normal end of iteration, so mapiter_set/mapiter_get returned success with the exception still set. depending on what runs next that's either a silent bogus success (the error surfaces at a later unrelated call) or SystemError: ... returned a result with an exception set (e.g. via np.put_along_axis).

this adds a PyErr_Occurred() check after the iteration loops in mapiter_@name@. small arrays already raise cleanly since first-chunk failures are caught at NpyIter_Reset so this just makes big arrays consistent. kept the check unconditional (not needs_api-gated) because non-object casts can fail too, e.g. a string -> int cast hits the same leak

AI Disclosure

Used Claude to double check my work and mae sure I didn't miss anything