IntegerArray docs follow-up

[chris-b1]

xref #21160

Make it clear that it is opt-in only, by passing/casting dtype. For example [24] and [27] are still float64

More clear separation from numpy dtype - worry that 'int64' vs 'Int64' will be especially confusing for new people? Consider a different name altogether? (NullableInt64?)

In [23]: s = pd.Series([1, 2, np.nan])

In [24]: s
Out[24]: 
0    1.0
1    2.0
2    NaN
dtype: float64

In [25]: s = pd.Series([1, 2, 3])

In [26]: s[2] = np.nan

In [27]: s
Out[27]: 
0    1.0
1    2.0
2    NaN
dtype: float64