Cascading drop index fails if one of the shard indexes being dropped does not exist

Replicate by:

Create a huge distributed table (at least 3 shards, 300gb) and put a load on the database.

drop index concurrently some_index;
^C to abort before it's finished (leaves some indexes hanging around).
drop index concurrently some_index;

The second drop index will complain
ERROR: index "some_index_102232" does not exist
CONTEXT: while executing command on citus2.mydomain.com

And will not run until you re-create the subset of indexes that got dropped by the first drop index concurrently.

Expected to happen: Missing shard indexes in a 'drop index' silently ignored (after all, we *want* them to be missing, duh).