feat(avatar): auto-show idle avatar on MCP connection / boot

[kisaragi-mochi]

Background

After flashing the firmware and starting the gateway, the LCD does not display the avatar until an MCP client explicitly calls set_avatar(face="idle"). Until that call lands, the screen stays on whatever it was showing previously (xiaozhi-esp32 setup UI, or the last frame from the previous session) and the device appears "asleep" even though the WebSocket is connected and tools are reachable.

Observation

• During PR feat(tts): add VOICEVOX engine + Opus pipeline (#70 PR2) #75 real-device verification, the screen stayed dark / unresponsive immediately after get_status reported connected:true and 19 firmware tools.
• Calling set_avatar(face="idle") once brought the avatar up and made the device look unmistakably "alive".
• A first-time user reading the README would not necessarily know to make this call, so the kit's first-impression boot path is "I'm connected but nothing visible happens" — easy to confuse with a wiring or boot bug.

Proposal

Pick one — (A) is recommended because it preserves the existing set_avatar(face="off") semantics for users who deliberately want the underlying xiaozhi screens (WiFi setup UI, OTA, etc.):

• A. Default after first MCP tools/list exchange. Once the gateway finishes initialising the device session, if no set_avatar has been called yet, the firmware renders idle. Subsequent set_avatar(face="off") continues to work as documented and is not overridden by this default.
• B. Default on boot. Render idle immediately after the WiFi/WS handshake completes, regardless of MCP state. Simpler to reason about, but removes the "no-MCP-yet, screen still dark" affordance for users who rely on it.

Out of scope

• Per-user customisation of the default face (config / NVS) — the question here is "does any face show by default", not "which one".
• Changing the existing set_avatar(face="off") semantics. The "off" face must remain reachable so the xiaozhi screens are still accessible.

Refs

• feat(tts): add VOICEVOX engine + Opus pipeline (#70 PR2) #75 — surfaced this UX gap during real-device verification
• Blink end repaints full face, overwriting active mouth overlay (Phase 2 trade-off, affects set_mouth and set_mouth_sequence) #65 — related avatar rendering concern (different subsystem detail, kept independent)