The snapshot error is not propagated to the initiating node if it is not the coordinator.

In some cases, when we execute "createSnapshot" on non-coordinator node, the snapshot error is not reported in the "LastSnapshotErrorMessage" metric, but the "LastSnapshotStartTime" and "LastSnapshotErrorMessage" metrics are updated.
This is misleading that the snapshot was taken successfully.

The problem concerns not only metrics, despite the error, the user future completes successfully and error does not reach the user.

Reproducer:

  startGridsWithCache(2, dfltCacheCfg, CACHE_KEYS_RANGE);

  // Must fail, but on non-coordinator finishes successfully.
  grid(1).context().cache().context().snapshotMgr().createSnapshot(SNAPSHOT_NAME, "/bad/path").get();

  // Fails (as expected).
  grid(0).context().cache().context().snapshotMgr().createSnapshot(SNAPSHOT_NAME, "/bad/path").get();