Salvage of #16620 by @julysir onto current main.

Summary

When replying inside a Feishu topic-enabled group chat, the agent created a new topic/thread instead of staying in the current one. The reply anchor was hard-set to the CURRENT message_id rather than the incoming reply's parent. On Feishu, with a thread_id present AND a reply_to_message_id available, use reply_to_message_id as the anchor so the agent's reply lands inside the existing thread. Also extend _process_inbound_message to fall back to root_id when computing thread_id / reply_to_message_id so Feishu's Topic Mode API shape is handled.

Conflict resolution during salvage

Main has since added ephemeral-message handling (_unwrap_ephemeral, _schedule_ephemeral_delete) at the two dispatch call sites this PR targets. Re-applied the Feishu reply-anchor fix in-place on the HEAD send calls (patching reply_to= inside the existing _r = await self._send_with_retry(...) blocks) rather than re-introducing the PR's parallel send block which would now double-send the response.

Changes

• gateway/platforms/base.py: Feishu+thread_id → use reply_to_message_id as reply anchor at 3 send sites (+32/-4 net)
• gateway/platforms/feishu.py: fall back to root_id for thread_id / reply_to_message_id in _process_inbound_message

Validation

scripts/run_tests.sh tests/gateway/ -k feishu -> 377 passed

Original PR: #16620