fix(desktop): ship the VC++ runtime the Windows voice wheels need

onnxruntime (wake word) and ctranslate2 (local speech-to-text) link against
msvcp140.dll and msvcp140_1.dll, which the managed CPython does not ship.
On Windows without the system VC++ Redistributable both imports failed with
"DLL load failed: The specified module could not be found", so "Hey Hermes"
and voice chat transcription never worked.

- Bundle the Microsoft-signed VC143 CRT set (14.44.35211, matching the
  managed CPython's vcruntime140) as a Windows extra resource with a SHA-256
  manifest.
- Before every local backend start, copy missing or mismatching CRT files
  beside the managed python.exe (the process image directory, searched first),
  verifying digests; never overwrite the interpreter's own vcruntime and never
  block boot. Existing installs heal on their next launch.
- Report an unloadable native voice wheel as a VC++ runtime problem in the
  wake-word requirements, local transcription and `hermes doctor`, instead of
  treating installed package metadata as proof the wheel loads.
- Bump AITURK IDE to 1.0.0-beta.12.

Co-Authored-By: Claude Opus 5 <noreply@anthropic.com>

tools/transcription_tools.py

@@ -1942,6 +1942,14 @@ def _transcribe_local(
         if not _try_lazy_install_stt():
             return {"success": False, "transcript": "", "error": "faster-whisper not installed"}
 
+    # faster-whisper imports lazily, so an unloadable ctranslate2 (missing
+    # VC++ runtime on Windows) would otherwise surface as an opaque DLL error.
+    from tools.windows_native_runtime import native_load_error, voice_runtime_error_message
+
+    native_error = native_load_error("ctranslate2")
+    if native_error:
+        return {"success": False, "transcript": "", "error": voice_runtime_error_message(native_error)}
+
     try:
         local_cfg = _load_stt_config().get("local") or {}
         # Reset the idle timer BEFORE loading/transcribing so the idle-unload

tools/wake_word.py

@@ -896,6 +896,12 @@ def _tts_ready() -> bool:
         return False
 
 
+def _native_runtime_error(module: str) -> Optional[str]:
+    from tools.windows_native_runtime import native_load_error
+
+    return native_load_error(module)
+
+
 def check_wake_word_requirements(cfg: Optional[Dict[str, Any]] = None) -> Dict[str, Any]:
     """Report whether wake-word detection can run, with a remediation hint."""
     cfg = cfg if cfg is not None else load_wake_word_config()
@@ -910,6 +916,10 @@ def check_wake_word_requirements(cfg: Optional[Dict[str, Any]] = None) -> Dict[s
         feature = "wake.openwakeword"
     deps_ok = lazy_deps.is_available(feature)
     lazy_ok = lazy_deps._allow_lazy_installs()
+    # Installed metadata does not prove the native wheel loads: on Windows
+    # without the VC++ runtime onnxruntime is "installed" yet every import
+    # fails. Report that instead of a lazy install that cannot help.
+    native_error = _native_runtime_error("onnxruntime") if deps_ok and feature == "wake.openwakeword" else None
     # The audio probe imports sounddevice + numpy — two of the very packages
     # the lazy installer would fetch — so it can only be trusted once the
     # feature's deps are installed. On a fresh install (deps missing, lazy
@@ -937,6 +947,10 @@ def check_wake_word_requirements(cfg: Optional[Dict[str, Any]] = None) -> Dict[s
     if provider == "porcupine" and not (os.getenv("PORCUPINE_ACCESS_KEY") or "").strip():
         key_ok = False
         hint = "Set PORCUPINE_ACCESS_KEY (free key at https://console.picovoice.ai)."
+    elif native_error:
+        from tools.windows_native_runtime import voice_runtime_error_message
+
+        hint = voice_runtime_error_message(native_error)
     elif not deps_ok and not lazy_ok:
         hint = lazy_deps.feature_install_command(feature) or ""
     elif not tflite_ok:
@@ -969,7 +983,7 @@ def check_wake_word_requirements(cfg: Optional[Dict[str, Any]] = None) -> Dict[s
             )
 
     return {
-        "available": key_ok and stt_ok and tts_ok and tflite_ok and mic_ok,
+        "available": key_ok and stt_ok and tts_ok and tflite_ok and mic_ok and native_error is None,
         "provider": provider,
         "deps_available": deps_ok,
         "audio_available": audio_ok,
@@ -979,6 +993,7 @@ def check_wake_word_requirements(cfg: Optional[Dict[str, Any]] = None) -> Dict[s
         "stt_available": stt_ok,
         "tts_available": tts_ok,
         "phrase": wake_phrase(cfg),
+        "native_runtime_error": native_error,
         "hint": hint,
     }
 

tools/windows_native_runtime.py

@@ -0,0 +1,77 @@
+"""Detect native voice wheels that are installed but cannot load on Windows.
+
+Package metadata (what ``tools.lazy_deps`` checks) reports onnxruntime and
+ctranslate2 as installed even when their DLLs cannot load because the
+Microsoft Visual C++ runtime (msvcp140.dll) is missing. A lazy reinstall cannot
+fix that, so callers need to tell the two cases apart. Importing the module in
+a child interpreter does so without risking the calling process.
+"""
+
+from __future__ import annotations
+
+import functools
+import os
+import re
+import subprocess
+import sys
+from typing import Optional
+
+_MODULE_RE = re.compile(r"^[A-Za-z_][A-Za-z0-9_.]*$")
+
+# Windows reports the same loader failure in the user's UI language.
+_DLL_LOAD_MARKERS = (
+    "dll load failed",
+    "could not find module",
+    "specified module could not be found",
+    "belirtilen modül bulunamadı",
+    "[winerror 126]",
+)
+
+VC_RUNTIME_HINT = (
+    "Ses bileşeni yüklenemedi: Microsoft Visual C++ çalışma zamanı (msvcp140.dll) eksik. "
+    "AITURK IDE'yi kapatıp yeniden açın; IDE eksik dosyaları otomatik olarak geri yükler."
+)
+
+
+def looks_like_dll_load_failure(message: str) -> bool:
+    text = (message or "").lower()
+    return any(marker in text for marker in _DLL_LOAD_MARKERS)
+
+
+@functools.lru_cache(maxsize=None)
+def native_load_error(module: str) -> Optional[str]:
+    """Return the DLL load error for *module* on Windows, otherwise None.
+
+    None also covers "not installed" and inconclusive probes: this helper only
+    answers "installed but its native libraries cannot load". Cached per
+    process because the runtime cannot change under a running backend (the
+    desktop restores it before the backend starts).
+    """
+    if sys.platform != "win32":
+        return None
+    if not _MODULE_RE.match(module):
+        raise ValueError(f"invalid module name: {module!r}")
+
+    try:
+        proc = subprocess.run(
+            [sys.executable, "-c", f"import {module}"],
+            capture_output=True,
+            text=True,
+            encoding="utf-8",
+            errors="replace",
+            timeout=120,
+            env=dict(os.environ, PYTHONIOENCODING="utf-8"),
+            creationflags=getattr(subprocess, "CREATE_NO_WINDOW", 0),
+        )
+    except (OSError, subprocess.TimeoutExpired):
+        return None
+
+    stderr = proc.stderr or ""
+    if proc.returncode == 0 or not looks_like_dll_load_failure(stderr):
+        return None
+    lines = [line.strip() for line in stderr.splitlines() if line.strip()]
+    return lines[-1] if lines else f"import {module} failed"
+
+
+def voice_runtime_error_message(error: str) -> str:
+    return f"{VC_RUNTIME_HINT} ({error})"