Standardize pandas metadata for table schema and parquet

[TomAugspurger]

Both to_json(orient='table') and to_parquet write some metadata about the dataframe. It'd be nice to standardize these.

Discrepancies

• Default index names ('index' vs. '__index_level_d__')

We should also add a function for parsing the metadata, and perhaps constructing an empty DataFrame with the correct index, types, and column names based off the metadata.

[WillAyd]

So are you thinking of porting some of the code from arrow over into pandas to accomplish this? Haven't used parquet before but I see that the below point in the pandas_compat module is where the index names you mentioned are appearing.

https://github.com/apache/arrow/blob/422efd9635ea6f249adec7e1fda4834f6ac46cc4/python/pyarrow/pandas_compat.py#L182