Clean up reshaping ops

[mroeschke]

Description

Uses some more "idiomatic" cudf patterns such as

• Checking isinstance(column.dtype, ...) instead of isinstance(column, ...) (to avoid importing the column objects)
• Using DataFrame._from_data(dict) instead of creating an empty DataFrame and adding columns one by one

Also avoids some column materialization in DataFrame.columns = :

• For RangeIndex, avoid materializing to a column to get a distinct count
• For MultiIndex, avoid creating a cudf.MultiIndex with columns as it's converted to a CPU object to get column labels for the ColumnAccessor

Checklist

• I am familiar with the Contributing Guidelines.
• New or existing tests cover these changes.
• The documentation is up to date with these changes.

[vyasr]

/merge