MasterServer.Shutdown only stopped the Hashicorp raft implementation;
when using the default goraft backend, the raft event-loop goroutine
(leaderLoop/followerLoop) kept running after the master shut down. In
the in-process test harness this leaked goroutines across sequential
test runs, and a stale event occasionally reached a leader at term 0
and tripped the goraft "leader.elected.at.same.term" assertion,
crashing the whole test binary (CI run 34670959967, PR 11279).
Stop the goraft server in Shutdown() so its goroutines exit cleanly,
and bump seaweedfs/raft to v1.2.1 which replaces that assertion with a
graceful step-down to Follower instead of a panic.
Chris Lu
c46f82d29a