Fix node-removal state handling and pause/resume orchestration

[Hamdy-khader]

Summary

This PR fixes storage-node removal state handling during node drain/removal workflows and makes the removal lifecycle explicit and resumable.

What changed

• Added explicit node states for removal progress:

  • pending_removal
  • in_removal

• Updated distribution/status reporting to treat these states as non-schedulable/unreachable so they are not counted as healthy cluster members.

• Updated node-removal orchestration to:

  • mark a node as pending_removal before the removal task becomes active
  • transition to in_removal once the orchestration begins
  • resume cleanly on retries without re-running completed phases
  • keep “retry later” behavior when peer nodes are still unavailable or migration work is incomplete

• Preserved correct phase ordering for:

  • graceful shutdown
  • replica relocation
  • finalization
  • device decommissioning