[Fix]: cm handle DEAD-rejoin and STANDBY lifecycle to prevent false reshard triggers

[ChillyAokiji]

Fix CM node lifecycle: DEAD rejoin and STANDBY state handling

When a DS restarts after being marked DEAD (i.e., the deferred reshard window had already expired and shards were redistributed), the CM previously set the node back
to RUNNING. This caused two problems:

• False HB timeout loop: the rejoining DS had stale (expired) heartbeat records, so the HB scan immediately re-entered DEFERRED_RESHARD on the next scan cycle.
• Inflated alive count: the empty DS was counted as a valid RUNNING node, skewing the reshard feasibility check (alive >= dataserver_min_num) and potentially
  receiving shards during rebalance despite holding no data.

Fix: introduce a STANDBY state for post-DEAD rejoins. A STANDBY node is tracked but excluded from alive counts and shard rebalance targets. If it goes silent again,
HB timeout transitions it directly to DEAD with no reshard (it holds no shards). A new unit test covers the full RUNNING → DEFERRED_RESHARD → DEAD → STANDBY → DEAD
lifecycle.

[Sebastian-Keith]

LGTM