Support and validate "not applicable" in CSV metadata parsing

[brianhelba]

The value "not applicable" should be accepted by the metadata parser, and should ultimately result in a null value being emitted as the field value.

However, as a guard against mistakes from the metadata contributor, the inter-field validation step should also ensure that values marked "not applicable" are truly not applicable in the given metadata context (and not accidentally used in places where "unknown" would be semantically more appropriate).

This will likely require the implementation to temporarily differentiate between "unknown"/"" values and "not applicable" values, while still saving both as None in the "load" step.