Hide value related GRIB metadata in NumpyFieldList

[sandorkertesz]

Current situation

If we create a new NumpyFieldList with the following code:

# f is a GribField
vals_new = f.values + 1
md_new = f.metadata().override(shortName="pt")
ds_new = FieldList.from_numpy(vals_new, md_new)

then the each new NumpyField consists of a numpy array (vals_new) and a metadata object, which stores a GRIB handle cloned from the handle in the original GRIB field but the values in it are not updated with vals_new. See #203 for an explanation.

As a result, there is an inconsistency between the various Field methods related to the values:

# these use the numpy array
ds_new[0].values
ds_new[0].to_numpy()
ds_new[0].data()

# these use the metadata object
ds_new[0].metadata("max")
ds_new[0].metadata("bitmapPresent")
ds_new[0].metadata("numberOfMissing")
ds_new[0].metadata(namespace="statistics")
ds_new[0].dump()
ds_new[0].dump(namespace="statistics")

>>> ds_new[0].values.max()
234.12
>>> ds_new[0].metadata("max")
233.12

The proposed solution

These GRIB keys are in fact not metadata keys but rather related to the actual values in the field. Moreover we can get the correct value related stats by inspecting the numpy array. So the right decision is to hide them in a NumpyFieldLists.

The expected behaviour:

# this throws a KeyError
ds_new[0].metadata("max")
ds_new[0].metadata()["max"]

# these return None
ds_new[0].metadata("max", None)
ds_new[0].metadata().get("max")

The "statistics" namespace will be hidden, whether we use the namspace kwarg in metadata() or we use the dump(). E.g. this is how the dump() output would look like: