Mixed types in ferc2 partition value causes complexity downstream

[zaneselvans]

The early ferc2 data was split into two parts per year (split alphabetically based on the name of the respondent) while in later years the data is not split. There are some years in which both split and unsplit data were published.

We use a partition named part to identify these portions of the data, with values of: 1, 2, or None.

The use of mixed types in the partition values creates some probably unnecessary complexity downstream, where the different values need to be handled separately. (e.g. you can't sort the partition values, because None can't be compared with an integer)

If possible we should probably try and keep the partition values homogeneous in type to avoid this complexity. In this case, maybe we could use part=0 to indicate that the data is unsplit?