dropna on a pyarrow float subcolumn silently keeps NaN rows

NestedFrame.dropna(subset=["nested.field"]) does not drop NaN when the subcolumn is pyarrow-backed and the missing values are float NaN rather than Arrow nulls.
Reported by @drewoldag in https://discovery-alliance.slack.com/archives/C04610PQW9F/p1785974842086829, when they tried filtering tess_lightcurve on lightcurve.sap_flux with many NaN values which were not removed.

import numpy as np
import pandas as pd
import pyarrow as pa
from nested_pandas import NestedFrame

flux = pa.array([1.0, np.nan, 3.0, np.nan], type=pa.float64())
flat = pd.DataFrame({"flux": pd.arrays.ArrowExtensionArray(flux)}, index=[0, 0, 1, 1])
nf = NestedFrame({"id": [10, 11]}, index=[0, 1]).join_nested(flat, "lc") # a nested float column holding NaN

print(len(nf.dropna(subset=["lc.flux"])["lc"].nest.to_flat())) # 4, expected 2

Start by running the provided NestedFrame.dropna(subset=["lc.flux"]) reproduction and inspect the dropna path for nested pyarrow-backed subcolumns. The fix is done when float NaN rows are removed as expected while the resulting nested data retains the two valid rows; verify the example output after the change.