Fix flaky TestShardController_SwapNodeWithLeaderElectionFailure

After an election, followerCaughtUp is set asynchronously in a background goroutine (ensureFollowerCaught at shard_controller_election.go:513-515), but SteadyState is set synchronously when start() returns. The test waited for SteadyState and then called ChangeEnsemble, which could be rejected with ErrNotReadyForChangeEnsemble if the background goroutine hadn't finished yet.

The error was silently swallowed because wg.Wait() was never called in the test, and the test hung for 10 seconds waiting for NewTerm requests that were never sent.

Proof by elimination:

• The test fails because expectNewTermRequest times out — no NewTerm RPC was sent
• onChangeEnsemble has only two code paths:
  1. IsReadyForChangeEnsemble() returns false → return ErrNotReadyForChangeEnsemble (no NewTerm sent)
  2. Call onElectLeader → NewTerm is sent
• Since no NewTerm was sent, path 1 was taken
• IsReadyForChangeEnsemble() returns e.followerCaughtUp.Load(), which is false until the background goroutine completes

Fix:

• Test: Retry ChangeEnsemble in a loop until it's accepted (not rejected with ErrNotReadyForChangeEnsemble). Applied to both TestShardController_SwapNodeWithLeaderElectionFailure and TestShardController_LeaderElectionShouldNotFailIfRemoveFails.
• Production code: Added a warning log when the rejection happens, for observability.
• Mock helpers: Added return after assert.Fail() in all 6 expect*Request functions to prevent nil pointer panics when the underlying timeout fires.