Index manager initialization may be costly, in particular in the Elasticsearch case, and may not involve concurrent access to shared resources.
We might want to initialize index managers in parallel, in order to improve performance.
Note: this may conflict with the idea of lazy index manager initialization from HSEARCH-2339.
Conversely, closing an index manager involves switching off work sinks (not accepting any more work) and then waiting for pending works to finish executing. The second part should definitely be executed in parallel for all index managers.