[BUG] use_arrow_dtype doesn't work when read a directory path for read_parquet

[hekaisheng]

Describe the bug
use_arrow_dtype doesn't work when read a directory path for read_parquet

To Reproduce

In [26]: df = md.read_parquet(tempdir, use_arrow_dtype=True)

In [27]: df.dtypes
Out[27]: 
a      int64
b     object
c    float64
dtype: object

Expected dtypes should be:

In [14]: df.dtypes
Out[14]: 
a            int64
b    Arrow[string]
c          float64
dtype: object

Expected behavior
A clear and concise description of what you expected to happen.

Additional context
Add any other context about the problem here.