fix(ui): prevent chat rail flicker when switching conversations

[vedantlavale]

Closes #3078

What changed

When a reader switches conversations, the route contains only the conversation ID while the selected instance is loading. The rail now resolves the Agent from the already-loaded conversation list during that gap, so the Agent identity, sibling filter, Agent Details link, and New chat link remain stable.

The transcript also keeps its loading skeleton visible until the selected conversation record is ready, instead of briefly rendering an empty conversation.

How it was tested

• NODE_OPTIONS=--localstorage-file=/tmp/kagent-ui-test-storage corepack yarn vitest run src/components/agent/AgentRail.loading.test.tsx — 4 tests passed.
• corepack yarn typecheck — passed.
• Targeted ESLint for the changed UI and Playwright files — passed.
• git diff --check — passed.
• Manual mock UI verification with ?mock=slow while switching from the seeded ready conversation to its sibling.

Before and after

Captured from the same route and mock scenario. The before capture uses local main; the after capture uses this branch.

Before	After

Risks and follow-ups

This is a UI-only change. It does not change the API, persistence, or conversation lifecycle. The Playwright browser executable was unavailable in the local environment, so the focused browser test remains for CI and manual verification.

[Charlesthebird]

This automated review is a first pass. A manual review will follow once the blockers are addressed.

---

Nothing blocks merge. The fix is correct, and the before/after recording in the next comment shows the rail and transcript stay stable during a switch.

✅ What I checked

• ✅ List entry vs. instance.agent: both map through toAgentInstance, and the list reads every page. A conversation missing from the list falls back to the base behaviour.
• ✅ isLoading={instance.isLoading}: true only on first load for a key. Background re-reads, polling and errors don't flash or stick the skeleton (shouldRetryOnError: false).
• ✅ ChatTranscript keeps its hooks above the early return, and the composer lives in AgentChatPage, so the skeleton loses no state.
• ✅ The new unit test fails on base; the new Playwright test fails on base only at its skeleton check (agent-rail.spec.ts#L174).

---

🤖 written by Claude

[Charlesthebird]

Before/after of switching to a sibling chat with ?mock=slow. Before, the rail shows the raw ID and every chat, and the transcript shows an empty state. After, the rail keeps the agent and the transcript shows a skeleton.

pr3218-reel.mp4

---

🤖 written by Claude

[Charlesthebird]

Thanks for adding this! I think removing the test would be fine for this and then I can approve.