Add operation to replace missing values

[dlindhol]

This relies on missingValue metadata. If there are no missing values defined in the metadata treat this as a no-op.

[dlindhol]

There can also be a fillValue in the metadata (#146). The ReplaceMissing operation should replace values that match either missingValue or fillValue. The metadata should be updated with missingValue set to the new value and fillValue should be removed.

Note that "NaN" and "null" require special logic beyond simple equality.

Note that a dataset with NullData should have missingValue="null" in the metadata. A fillValue in the metadata must not be "null".

[dlindhol]

ReplaceMissing needs to be targeted to a specific variable since the data types may be incompatible.
Consider ReplaceAllMissing to try to replace for all variables. This is a common scenario (as implemented in latis2's ReplaceMissingOperation) where all variable with missing values are doubles.

Note that latis3 needs to be able to validate the applicability of an operation based on metadata and model alone.