In this simple initial implementation we are limited in how quickly we detect
a failure by a worst case of roughly zk session timeout + WAIT_AFTER_EXPIRATION_SECONDS + WORK_LOOP_DELAY_MS
and best case of roughly zk session timeout + WAIT_AFTER_EXPIRATION_SECONDS. Also, consider the time to
create the SolrCore, do any recovery necessary, and warm up the readers.
NOTE: this will only work with collections created via the collections api because they will have defined
replicationFactor and maxShardsPerNode.
WARNING: This API is experimental and might change in incompatible ways in the next release.