Summary

• Add MiMo-V2.5-ASR provider support for Xiaomi's speech recognition API
• Uses OpenAI-compatible chat completions endpoint with base64 audio input and SSE streaming
• Follows the same architecture as the existing GLM provider (buffer audio → POST on finish → parse SSE stream)

Changes

New file:

• koe-asr/src/mimo.rs — MiMo ASR provider implementation with 19 unit tests (WAV wrapping, request body construction, response parsing, SSE line parsing, provider lifecycle)

Modified files:

• koe-asr/src/lib.rs — register mimo module and re-export MimoAsrProvider
• koe-core/src/config.rs — add MimoAsrConfig struct with url, api_key, model, language fields, defaults, and YAML config entry
• koe-core/src/lib.rs — add match arm for "mimo" provider selection
• KoeApp/.../SPSetupWizardWindowController.m — add MiMo to setup wizard (popup item, API key field with secure/plain toggle, show/hide logic, config save/load, test connection method, pane height)

How it works

1. Audio PCM is buffered during send_audio()
2. On finish_input(), PCM is wrapped in WAV, base64-encoded, and sent as data:audio/wav;base64,... in a chat completions request
3. SSE streaming response is parsed for choices[0].delta.content (streaming) or choices[0].message.content (non-streaming)
4. Authentication via api-key header

Config example

asr:
  provider: "mimo"
  mimo:
    api_key: "your-key"
    model: "mimo-v2.5-asr"
    language: "auto"

Test plan

• 82/82 Rust tests pass
• cargo fmt --check clean
• cargo clippy — no new warnings
• Manual test: configure MiMo provider in setup wizard
• Manual test: record audio and verify transcription
• Manual test: test connection button with valid/invalid API key