fix(web): finish local prompt state from session snapshot after a reconnect

[wbxl2000]

Related Issue

No linked issue — the problem is explained below.

Problem

In the web UI, a turn that finishes while the client is disconnected (laptop
closed, backgrounded tab, network drop) can leave the session stuck: the
working spinner never stops and the next message is queued behind a turn that
already ended.

The missed turn end is normally recovered by durable-event replay on
reconnect, but not when the client rebuilds from a session snapshot instead —
server-initiated resync (buffer overflow / epoch change / delta gap) or an
LRU-evicted subscription rebuild. Local prompt state (in-flight flag, sending
flag, cached prompt id, the message queue) was only ever cleared by the WS
sessionStatusChanged idle/aborted event, and snapshots never delivered one.

What changed

Web-only fix; no server, protocol, or UI style changes.

• A shared local-turn-start lifecycle (generation + submit-pending, in
  useWorkspaceState) is now entered by every local turn start: prompt
  submit, steer, and skill activation. A snapshot requested before or during
  a turn start can tell it cannot reflect that turn.
• syncSessionFromSnapshot rejects a snapshot that predates a local turn
  (or raced one) and re-syncs once the turn start settles, instead of
  permanently ignoring it. Snapshots that fall behind the live cursor during
  the fetch are dropped (one retry) rather than applied — resync/LRU rebuilds
  still always apply because their projector was deliberately reset.
• When an authoritative snapshot shows the turn ended (inFlightTurn === null, status idle/aborted), it finishes the local prompt state through the
  same cleanup the WS idle path uses (finishPromptLocal), without firing
  completion notifications/sounds — opening a historical session must not cry
  wolf.
• Turn-end side effects on the WS path are gated on the durable cursor
  actually advancing, so a late duplicate idle can never drain a second
  queued message.
• window focus / online also trigger the existing stale-socket recovery
  (previously only visibilitychange); recovery still only reconnects when
  the socket actually looks stale.

Tests cover the snapshot finish path, the running/awaiting cases, the
new-prompt-during-fetch and submit-pending rejections, and the settle
callback retry.

Checklist

• I have read the CONTRIBUTING document.
• I have linked a related issue, or explained the problem above.
• I have added tests that prove my feature works.
• Ran gen-changesets skill, or this PR needs no changeset.
• Ran gen-docs skill, or this PR needs no doc update.

[changeset-bot]

🦋 Changeset detected

Latest commit: c8ed644

The changes in this PR will be included in the next version bump.

This PR includes changesets to release 1 package

Name	Type
@moonshot-ai/kimi-code	Patch

Not sure what this means? Click here to learn what changesets are.

Click here if you're a maintainer who wants to add another changeset to this PR

[pkg-pr-new]

pnpm dlx https://pkg.pr.new/@moonshot-ai/kimi-code@c8ed644

npx https://pkg.pr.new/@moonshot-ai/kimi-code@c8ed644

commit: c8ed644