Fix primary relocation for shadow replicas #22474
Primary relocation for shadow replicas is broken as the recovery process accesses the engine on the source shard after it's been closed. This results in the source shard failing itself. This has not been detected by tests as, after failing source and target shard of the relocation, the master simply reassigns the primary to the node where it wanted to relocate to as every node has access to the shard files.
Test failure exhibiting behavior (might be an un-related issue though):
Relates to #20300