PERF/CLN: restrict _isna_ndarraylike to actual ndarray/EA

[jorisvandenbossche]

xref #39146

I am looking at some isna related benchmarks, and noticing a lot of overhead related to checking of types. This makes _isna_ndarraylike stricter to only handle ndarray and EA, and not Series/Index (and therefore also renamed it to _isna_array).

Reproducer from the frame_methods.py::Isnull benchmark:

N = 10 ** 3
sample = np.array([np.nan, 1.0])
data = np.random.choice(sample, (N, N))
df = DataFrame(data)
df_am = df._as_manager("array")

In [3]: %timeit df_am.isna()
5.03 ms ± 125 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)  <-- master
4.25 ms ± 9.94 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)  <-- PR

This gives a small improvement. But it further allows to directly use _is_array instead of the general isna from Manager.isna, which will give further improvements (but was planning to do this as a follow-up)

[jreback]

lgtm

[jbrockmendel]

LGTM pending green