Describe the problem

Raft snapshots are occasionally sent in healthy systems. They should only be required in the case of failures or slowness of nodes.

To Reproduce

(Note this test is included in the PR)

• Create a cluster with 3 nodes and the raft snapshot queue disabled.
• Create a range on just 1 node.
• Add a second node as a replica.
• Have the replicate queue attempt to add the third node (since it doesn't like an even number of replicas)

Expected behavior
Most of the time this succeeds with the third node added. Occasionally it will fail.

Jira issue: CRDB-19406