There are several concurrency tasks that have long-running times. We should be able to dynamically split these to bring the wall-clock time down.

Also, these require running on large distros to avoid OOM, but new build variants are being added that don't run them on the right distros. By splitting them dynamically, we can handle that without new build variants needing to know.

The tasks we will split:

• concurrency_replication_causal_consistency
• concurrency_sharded_causal_consistency_and_balancer
• concurrency_sharded_clusterwide_ops_add_remove_shards
• concurrency_sharded_local_read_write_multi_stmt_txn
• concurrency_sharded_local_read_write_multi_stmt_txn_with_balancer
• concurrency_sharded_multi_stmt_txn
• concurrency_sharded_multi_stmt_txn_with_balancer
• concurrency_sharded_multi_stmt_txn_with_stepdowns
• concurrency_sharded_replication
• concurrency_sharded_replication_no_txns
• concurrency_sharded_replication_no_txns_with_balancer
• concurrency_sharded_replication_with_balancer
• concurrency_sharded_with_stepdowns
• concurrency_sharded_with_stepdowns_and_balancer