Ensure categorical columns use appropriate NA values

[zaneselvans]

In some of our categorical columns, at the end of the transform step we have only the enumerated values, plus the empty string. In the database, these columns are nullable, and so the empty string values should really be pd.NA. However, currently we have the empty string in some of our enumerations to avoid failing constraint checks. These should all be removed, and the corresponding columns should have the empty string replaced with pd.NA before loading. Columns that this is happening in include:

FERC 1

• construction_type

EIA

• natural_gas_delivery_contract_type_code
• natural_gas_transport_code
• purchase_type

EPA CEMS

• co2_mass_measurement_code
• nox_mass_measurement_code
• nox_rate_measurement_code
• so2_mass_measurement_code

[ezwelty]

I listed additional placeholder values that need to be replaced with null in https://github.com/catalyst-cooperative/pudl/pull/806#issue-509727837 and removed in the metadata, namely:

• [ ] Remove "unknown" from constraints.enum for fields:
  • construction_type
  • fuel_unit
• [ ] Remove "U" / "Unknown" from constant ENTITY_TYPES.