Introduce explicit endpoint replacement flow for existing host_id

[dkropachev]

Parent epic: #921

Problem

The driver currently mutates Host.endpoint during topology refresh when a node keeps the same host_id but changes address. Direct mutation makes it hard to keep metadata indexes, session pools, load-balancing policies, client routes, and async connection work consistent.

Before Host.endpoint can be frozen, endpoint movement needs one explicit internal flow that owns the full transition.

Desired behavior

A same-host_id, different-endpoint observation should go through a single endpoint replacement/update path that:

• validates that the host_id is unchanged;
• updates metadata endpoint indexes atomically from the driver's perspective;
• retires or renews affected session pools;
• notifies load-balancing policies with clear remove/add or update semantics;
• invalidates/fences stale async pool and connection work;
• keeps runtime health behavior well-defined during the transition;
• preserves compatibility for public address lookup APIs where possible.

Scope

• Define the internal API for endpoint replacement.
• Route control connection topology refresh through that API.
• Ensure metadata, sessions, policies, and route state observe a coherent transition.
• Decide whether the existing Host object is retained with controlled internal update or replaced by a new Host object during the transition.
• Document the chosen semantics in code/tests.

Acceptance criteria

• There is one internal path for same-host_id endpoint movement.
• Callers no longer assign host.endpoint directly during topology refresh.
• Metadata endpoint indexes cannot retain stale endpoint mappings after replacement.
• Sessions do not keep using pools opened for the previous endpoint after replacement.
• Tests cover endpoint movement with preserved host_id.

Part of #921. Depends on #922 and #867. Related to #857, #858, #859, and #860.