BUG: fix ArrowExtensionArray constructor raising on np.nan in string sequences

[tinezivic]

Problem description

When constructing an ArrowDtype string array via pd.array() with a sequence that contains np.nan, an ArrowTypeError is raised:

>>> import pandas as pd
>>> import numpy as np
>>> pd.array(["a", np.nan], dtype=pd.ArrowDtype(pa.string()))
# ArrowTypeError: Expected bytes, got a 'float' object

Root cause

In ArrowExtensionArray._box_pa_array(), the mask is computed via:

arr_value = np.asarray(value)
mask = isna(arr_value)

np.asarray(["a", np.nan]) produces a <U3 Unicode array in which np.nan is silently coerced to the string 'nan'. isna() then returns [False, False], the NA mask is empty, and pyarrow rejects the float value.

Fix

Pass dtype=object to np.asarray() so that scalar types are preserved and isna() correctly identifies np.nan as a missing value.

Closes #64578