[FEATURE] Dynamic real-time voice switching in voice mode (spoken command + programmable per-session voices)

### Feature Request

Let users change Claude's voice in real time instead of it being one global app setting.

**The main hook is simple.** I want to just say "switch to a southern accent" or "sound like an older woman" mid-conversation and have it change right then. No trip to settings. That alone would be a killer demo.

**The deeper version is making that switch programmable so the user never has to ask.** I run multiple Claude Code sessions across several machines and use voice mode constantly. Voice should be a live runtime parameter. A session could declare its own voice in its config (CLAUDE.md or session metadata), or the user could build a lookup table that maps projects and topics to voices, and the rule fires automatically whenever that context comes up. For example, a hardware project always speaks in one voice, and the moment the conversation turns to stocks it flips to an older woman's voice because I set that rule once and never touched it again.

### Why this matters

When you are managing several agents by voice, giving each one a distinct voice makes them instantly identifiable. It stops feeling like a bunch of terminals and starts feeling like a team. Huge fun factor and a real differentiator.

You do not need voice cloning for any of this. The preset voice catalog you already have is plenty, it just needs to be exposed as something that can change on the fly and by spoken request instead of a single static toggle.

### Current behavior

Voice is a single static setting (Settings > Voice). It cannot change mid-conversation, per session, or by spoken request.

### Desired behavior

- A spoken command changes the voice live, mid-conversation
- Voice exposed as a per-session runtime parameter, settable via session config or metadata
- Optional user-defined mapping of projects and topics to voices, applied automatically

Related: #31724 (voiceLanguage setting) is the closest existing request, but nothing covers voice identity switching.

Start by reading the existing voice mode and Settings > Voice implementation, plus related issue #31724 for nearby voice configuration behavior. The requested end state is live voice switching by spoken command, a per-session runtime voice setting via CLAUDE.md or session metadata, and optional project/topic-to-voice rules. This is broad product/design work rather than a localized starter change.

Do mô hình lập chỉ mục viết ra từ nội dung của issue.