feat(settings): show Voxtype runtime controls - #49

Summary

• read Voxtype extended status and versioned config-schema JSON
• show CLI/daemon version, state, engine, model, audio device, compute backend, schema version, and config path in Settings
• launch Voxtype's own configure TUI in an available terminal instead of rewriting its configuration
• keep errors visible inside Settings and cover JSON/terminal selection with deterministic tests

Dependency

Depends on #48.

Validation

• PYTHONPATH=src /Users/bytedance/personal/doubao-say/.venv/bin/python -m unittest tests.unit.test_voxtype_control tests.unit.test_voxtype_runtime tests.unit.test_voxtype_asr_client tests.unit.test_recognition_providers
• /Users/bytedance/personal/doubao-say/.venv/bin/python -m ruff check src tests packaging
• PYTHONPATH=src /Users/bytedance/personal/doubao-say/.venv/bin/python -m compileall -q src tests
• audited Voxtype v1.0.1 status --extended --format json and config schema --json source contracts
• git diff --check

Validation gaps

• Voxtype 1.0.1 daemon was tested on Linux through a virtual PipeWire microphone; terminal configurator was not exercised
• no Linux GTK screenshot was captured; the UI still needs visual acceptance
• this PR intentionally delegates edits to voxtype configure; it does not embed a second configuration editor

Review order

Review after #48. This is the final PR in the recognition-provider stack.

Review fix (2026-09-25)

• Read Voxtype status on a worker thread and deliver results through GLib.idle_add, keeping GTK responsive while CLI commands run.
• Ignore stale status responses after the selected provider changes or a newer refresh starts. CLI timeout errors remain visible in Settings.
• Reproduced a two-second Settings freeze with a deterministic regression test before the fix; the same test passes afterward.
• Merged the feat(asr): integrate Voxtype local dictation #48 lifecycle fix into this branch so the PR diff still shows only the controls increment relative to its current base.

Validation on Linux: make PYTHON=/home/leyang/doubao-say/.venv/bin/python check and make PYTHON=/home/leyang/doubao-say/.venv/bin/python coverage passed (61% overall coverage). Voxtype 1.0.1 was subsequently tested with a virtual PipeWire microphone. A GTK screenshot and physical microphone acceptance were not captured.

Live Linux benchmark (2026-09-25)

• Merged the real-daemon command-runner fix from feat(asr): integrate Voxtype local dictation #48 (0631ad1); make check passes on this branch.
• With Voxtype 1.0.1, sensevoice-small ONNX CPU, and an isolated PipeWire source, release-to-app-result took 0.258–0.360 s across three full TranscriptionManager runs after capture became ready.
• feat(asr): integrate Voxtype local dictation #48 now documents the virtual-device cold-start diagnosis and waits for the daemon recording state before the app reports recording; this controls branch includes that fix.

Audio startup follow-up (2026-09-25)

• Merged feat(asr): integrate Voxtype local dictation #48 commit c866b8f so the app no longer reports recording as soon as Voxtype merely acknowledges the CLI signal. make check passes on this branch.
• The 3.2 s first-open measurement was specific to a newly created virtual PipeWire ALSA device. Three starts on the machine's default physical microphone reached the capture thread in about 25–35 ms. See feat(asr): integrate Voxtype local dictation #48 for the logs and live manager retest.