CSR_STATISTIC needs to be required

But just changing in the CSR_COLUMN_UNITS.csv file produces behavior we don't want:
```R
> csr_build(dataset_names = list_datasets()[1], quiet = TRUE)
Warning message:
In check_dataset_names(dsn, x, md) :
  d20190409_ANJILELI - required metadata fields not found for table 'ancillary': CSR_STATISTIC
```
Basically it should be required _if any ancillary data are present_.