Make it clear which node is causing the "shardsvr" error.

Spawned from support case 01042995

Our error occurred when the user tried to connect using Compass. The failure was to list the collection names on one database.

The error presented back to the user was merely Cannot accept sharding commands if not started with --shardsvr

We found eventually that the primary changed on one of the shards, and that primary did not have the appropriate clusterRole in the mongod.conf file. My concerns are that this took too long to track down and would be impossible in a 100-shard environment.

• Nothing in sh.status() indicated that there was any problem connecting to those nodes
• Nothing in the mongos.log file indicated which node was providing that error