validate_db_metadata_command.js fails in sharding_stepdown_fcv_upgrade_downgrade_jscore_passthrough because line 118 calls validate() immediately after dropIndexes() expecting zero API version errors, but a stepdown can cause validateDBMetadata to be routed to a newly elected primary that hasn't applied the drop yet (especially with catchUpTimeoutMillis: 0).

The test already retries with assert.soon when expecting errors to appear (assertValidateError), but lacks the symmetric retry when expecting errors to disappear.

Fix is to wrap the post-drop validate call in assert.soon so it retries until the clean state is visible across all shards.