Assistant text before a tool call (e.g. ask_user popup) is hidden in the collapsed "Thought for…" block instead of rendering as chat

Describe the bug

In interactive mode, any assistant chat text written in the same turn before a tool call (most notably the ask_user question popup) is never rendered as a visible chat message. It is instead folded into the collapsed "Thought for Ns" reasoning block, where it stays hidden unless manually expanded — even after the tool call resolves.

The practical impact: when the agent writes explanatory context ("here's the quoted text / options / proposal…") and then opens an ask_user popup referring to that context, the user only sees the popup and cannot see the text it refers to.

Affected version

1.0.73 (Copilot app / workspace UI; copilot --version in shell reports 26.5.2)

Steps to reproduce the behavior

Ran a series of controlled tests with the agent in interactive mode:

1. Agent emits a chat message containing a marker + blockquote, then calls ask_user in the same turn → user sees only the popup; the text never appears, even after answering. It is only findable by expanding the "Thought for Ns" block.
2. Agent emits a text-only turn (no tool calls) → text renders normally as visible chat.
3. Agent calls a tool first, then writes text after the tool result in the same turn → text renders normally.
4. Markdown inside the ask_user question field (blockquote, inline code, bold) renders correctly, and long question text is not truncated — so the popup itself is fine; only preceding same-turn text is lost.

Expected behavior

Assistant text preceding a tool call in the same turn should render as a normal, permanently visible chat message (before or alongside the popup), not be classified as model reasoning/thinking and collapsed.

Additional context

• Operating system: macOS (Darwin, ARM)
• Surface: Copilot desktop app (Tauri workspace UI) project session, interactive mode
• The misclassification appears to be in how the UI segments a turn: everything prior to the first tool call is treated as part of the thinking block rather than as assistant output.

首先，在交互模式下复现该问题：在调用 ask_user 之前先输出 assistant 文本；然后检查 workspace UI 如何围绕第一次工具调用对一轮进行分段。完成的标准是：之前的 assistant 文本会渲染为永久可见的聊天消息，同时 ask_user 弹窗继续正常工作。