term_full_entity_id has been added to wb_terms table to store the full entity ID serialization.
With https://gerrit.wikimedia.org/r/#/c/346169 making the code write to both "old" numeric id column and the "new" full ID column, it becomes possible to also use the new column for read operations.
As "full id" values for existing entries in wb_terms will be populated using maintenance script, initially database table would be in a transition state with some records having values in both entity ID column, and some having only the "old" numeric id value. Reading code should be able to handle such cases.
One way to handle this, as said in https://phabricator.wikimedia.org/T159851#3140376, would be to introduce three-state feature flag: 1) read only from numeric id column, 2) read only from the full id column, 3) read from the new column, if it does not exist or the value is null, read the numeric column.
As discussed with @daniel today, the alternative way would be to not add a a feature flag and instead always read all columns of wb_terms, and use the "full id" value if it is there, and fall back to the "numeric id" column if it there is no "full id" value or it is null.
As we talked with @daniel it might that in case of this schema change the latter approach might be good enough. It might make it easier without needing to deal with the extra burden introducing and then removing the feature flag brings.