Land improvments to the managewriter to improve connection management logic. This is impacting a particular customer workload.

Relevant internal issue: b/388522702

In consideration:

• reduce reconnects on schema changes
• improve draining for queued responses on a connection that's being recycled