Clone *mgo.Session (and close it within goroutine) to fix error with go test -race

[timvaillancourt]

Under go test with -race the TestUploadOplogToS3 panics. This seems to be due to the session being used in a goroutine that .Open() starts.

Without this PR, you get this panic with go test -race:

test_1       | === RUN   TestUploadOplogToS3
test_1       | panic: Session already closed
test_1       | 
test_1       | goroutine 53 [running]:
test_1       | github.com/globalsign/mgo.(*Session).cluster(0xc4200aa9c0, 0xb1f990)
test_1       | 	/go/src/github.com/globalsign/mgo/session.go:2071 +0xa5
test_1       | github.com/globalsign/mgo.(*Session).acquireSocket(0xc4200aa9c0, 0xa60701, 0x0, 0x0, 0x0)
test_1       | 	/go/src/github.com/globalsign/mgo/session.go:5141 +0x176
test_1       | github.com/globalsign/mgo.(*Database).Run(0xc4203d1d70, 0xa607a0, 0xc421ce0960, 0xa2a9c0, 0xc421cea000, 0x0, 0x0)
test_1       | 	/go/src/github.com/globalsign/mgo/session.go:992 +0x5e
test_1       | github.com/globalsign/mgo.(*Session).Run(0xc4200aa9c0, 0xa607a0, 0xc421ce0960, 0xa2a9c0, 0xc421cea000, 0xc4200ca1f0, 0xc4203d1de0)
test_1       | 	/go/src/github.com/globalsign/mgo/session.go:2489 +0xba
test_1       | github.com/percona/mongodb-backup/internal/cluster.GetIsMaster(0xc4200aa9c0, 0xc400000001, 0x1052, 0x2)
test_1       | 	/go/src/github.com/percona/mongodb-backup/internal/cluster/is_master.go:16 +0x1a3
test_1       | github.com/percona/mongodb-backup/internal/oplog.(*OplogTail).tailQuery(0xc4200ca180, 0x1)
test_1       | 	/go/src/github.com/percona/mongodb-backup/internal/oplog/oplog_tail.go:188 +0x549
test_1       | github.com/percona/mongodb-backup/internal/oplog.(*OplogTail).tail(0xc4200ca180)
test_1       | 	/go/src/github.com/percona/mongodb-backup/internal/oplog/oplog_tail.go:163 +0x4ce
test_1       | created by github.com/percona/mongodb-backup/internal/oplog.open
test_1       | 	/go/src/github.com/percona/mongodb-backup/internal/oplog/oplog_tail.go:86 +0x2b5
test_1       | FAIL	github.com/percona/mongodb-backup/internal/oplog	6.062s

After this PR I no longer get this error, but I'm not sure if the way I fixed this is ideal. @percona-csalguero: what do you think? It's also possible a 'defer' is triggering too early?

This change:

1. Passes a cloned *mgo.Session into the tailer goroutine .open().
2. The *OplogTail.Close() closes the cloned-session.