Ensure getLastErrorModes with replica set tags are satisfiable

This ticket is an extension to SERVER-18407 to cover custom write concern settings with replica set tags.

As with SERVER-18407, it is presently possible to create a custom write concern setting using replica set tags that is impossible to be satisfied. For example:

> rs.conf()

...

    "members": [

        {

            "_id": 0

            "tags": {

                "ssd": "installed"

            }

        },

        {

            "_id": 1

            "tags": {

                "ssd": "installed"

            }

        }

    ],

    "settings": {

        "getLastErrorModes": {

            "ssdWriteConcern": {

                "ssd": 2

            }

        }

    }

Please note the identical "ssd": "installed" tag defined in the replica set members.

Using this setting, attempts to insert into the collection using the user-defined custom write concern of w: ssdWriteConcern will always fail. The server apparently counts the number of unique values in the associated replica set tags to determine that the write concern was satisfied:

> db.test.insert({a:1},{writeConcern:{w:'ssdWriteConcern'}})

Inserted 1 record(s) in 3ms

WriteResult({

  "nInserted": 1,

  "writeConcernError": {

    "code": 100,

    "codeName": "CannotSatisfyWriteConcern",

    "errmsg": "Not enough nodes match write concern mode \"ssdWriteConcern\""

  }

})

The desired behaviour is to output a similar error message as per SERVER-18407 if the write concern can never be satisfied given a replica set configuration.