After SERVER-84340 introduced a new JS core suite triggering random step down events in the nodes of a ShardedClusterFixture, spurious test regressions have been observed, which may be resumed as follows:

• a sharded collection with hashed shard key gets created on a two shards cluster; each shard owns one chunk of the collection
• a coll.countDocuments({}) command (which resolves into an aggregation with 'primary' read preference) is issued through one of the mongos processes of the cluster
• between the establishment of the cursor and the consumption of the first results, the primary node of one shard is restarted, raising a CursorNotFound.

--> the error code gets unexpectedly returned to the test environment (instead of being remapped to a retriable one).