Right now, if you run mongodump --oplog against a standalone you get a pretty unhelpful error:

```
error getting oplog start: not found
```

When you run against a sharded cluster, you get a slightly more helpful error message, but still should be better IMO

```
error getting oplog start: can't use 'local' database through mongos
```

Might be worth telling the user to make sure they're running mongodump --oplog against a repl set in the "error getting oplog start" message