Reduce coordinator memory growth in citus_move_shard_placement

[Copilot]

DESCRIPTION: Bound coordinator memory use during shard moves with many colocations

citus_move_shard_placement() could retain large per-shard DDL structures on the coordinator, causing memory to climb dramatically for large colocated move sets (as in the attached WSS graph). This change makes the block-writes transfer path release transient allocations per shard instead of accumulating them for the full operation.

• What changed

  • Per-shard memory release in setup phase
    • Reset localContext during shard table recreation loop, so DDL/list allocations are freed each iteration.
  • Removed deferred command accumulation
    • Dropped ShardCommandList buffering of attach/FK DDL across all shards.
    • Execute partition-attach and foreign-key command lists incrementally per shard.
  • Ordering preserved
    • Keep operational ordering: partition attach work runs before foreign-key creation work.

• Illustrative change

SendCommandListToWorkerOutsideTransaction(targetNodeName, targetNodePort,
                                          tableOwner, ddlCommandList);
MemoryContextReset(localContext);