[FEA] Int64Index for header=None in CSV reader

[karthikeyann]

Is your feature request related to a problem? Please describe.
Pandas read_csv without header returns integers as column names. cudf read_csv returns integer numbers as strings.

import cudf
from io import StringIO
import pandas as pd
csv="1, 2, 3\n4, 5, 6\n7, 8, 9\n"

In [11]: cudf.read_csv(StringIO(csv), header=None).columns
Out[11]: Index(['0', '1', '2'], dtype='object')

In [12]: pd.read_csv(StringIO(csv), header=None).columns
Out[12]: Int64Index([0, 1, 2], dtype='int64')

Similar issue will happen in JSON reader also after "values" support in JSON reader PR #12498 (comment)

Describe the solution you'd like
The libcudf csv reader does not have a method to return metadata as non-string types. Provide a way to indicate there is no header information in metadata. It's possible to fix this issue at Cython/Python layer.

Additional context
PR Comment #12498 (comment)

[vuule]

@galipremsagar this looks like a Python feature. Is anything needed from libcudf to implement this?

[karthikeyann]

When header is not present, libcudf returns stringified integers as column names. The python layer may not know if the actual names are integers string, or they are auto-generated. So, libcudf may have to indicate somehow that column names are not present in metadata. That might be the change required from libcudf (It could simply leave the metadata.schema_info empty?)

[galipremsagar]

When header is not present, libcudf returns stringified integers as column names. The python layer may not know if the actual names are integers string, or they are auto-generated. So, libcudf may have to indicate somehow that column names are not present in metadata. That might be the change required from libcudf (It could simply leave the metadata.schema_info empty?)

+1