This deadlock has been caused by a combination of changes under the Authoritative Shards project (featureFlagAuthoritativeShardsDDL).
 
The test hangs because a chunk migration is stuck, its session migration destination thread cannot check out a session that is already held by ShardsvrSetAllowChunkOperationsCommand. That command in turn waits for a SplitChunkCoordinator to complete, but the coordinator cannot register itself in ActiveMigrationsRegistry because the very migration that is stuck holds the receive-chunk slot.

3-way deadlock on the recipient shard of a migration:

Chunk migration (Needs to checkout a 'migrated' session id) -> SetAllowChunkOperations (has checked out the same session id, needs to wait for split to finish) -> Split (need to acquire an ActiveMigrationsRegistry slot) -> Chunk Migration (holds the ActiveMigrationsRegistry slot)