Preserve manually typed text when voice mode sends in a new chat

[Copilot]

Description

In the Agents window, dictating into a new chat that already had manually typed text (plus an optional pasted image) sent a query containing only the voice-dictated portion — the typed text was dropped.

The backend voice LLM's send_to_chat tool routes the transcript through _chat.voice.acceptInput → NewChatInputWidget.sendQuery(text), which called model.setValue(text), overwriting the input before _send() read it. Attachments survived (they live in _contextAttachments); the typed text did not. The backend only knows the spoken phrase, so the typed portion must be preserved client-side.

• newChatInput.ts – sendQuery: append the dictated text to any existing input instead of replacing it.

const existing = model.getValue().trimEnd();
const combined = existing ? `${existing} ${text}` : text;
model.setValue(combined);
this._send();

Scoped to the composer path (the reported new-chat scenario); the existing-session widget.acceptInput flow is untouched. The _sending guard against double-submits from rapid transcripts is preserved.

[Copilot]

Pull request overview

Preserves manually entered draft text when voice input submits a new Agents window chat.

Changes:

• Appends dictated text to the existing draft.
• Retains the existing send guard and attachment handling.

Show a summary per file

File	Description
src/vs/sessions/contrib/chat/browser/newChatInput.ts	Combines existing draft and voice text before submission.

Review details

• Files reviewed: 1/1 changed files
• Comments generated: 3
• Review effort level: Medium