During the design of this project, there was one detail we were not aware. At every refresh in the sharding state, all instances of shard are re-created. This will cause the retry budget to be reset on all shards every time there's a topology change. This is not the desired behavior.

Instead, we should keep track of the shard retry budget per shard id, and send a pointer to the retry budget when constructing the shards.

At the same time, consider SERVER-109573 if possible. Since there will be a definite owner that will outlive the shards, we should be able to get rid of the shared pointers.