Use qd.interop.get_mps_command_queue() instead of inline extraction

The ctypes/ObjC logic to extract PyTorch MPS's MTLCommandQueue* now
lives in Quadrants (qd.interop module), so Genesis just calls the
public API. Removes ~35 lines of low-level ctypes code.

Co-authored-by: Cursor <cursoragent@cursor.com>

Author: hughperkins

genesis/__init__.py

@@ -1,4 +1,3 @@
-import ctypes
 import io
 import os
 import sys
@@ -53,41 +52,6 @@
 EPS: float | None = None
 
 
-def _get_mps_command_queue() -> int:
-    """Extract PyTorch MPS's MTLCommandQueue* as a Python int, or 0 on failure."""
-    try:
-        torch_lib = os.path.join(os.path.dirname(torch.__file__), "lib", "libtorch_cpu.dylib")
-        handle = ctypes.CDLL(torch_lib)._handle
-
-        libdl = ctypes.CDLL(None)
-        dlsym = libdl.dlsym
-        dlsym.restype = ctypes.c_void_p
-        dlsym.argtypes = [ctypes.c_void_p, ctypes.c_char_p]
-
-        stream_fn = dlsym(handle, b"_ZN2at3mps19getDefaultMPSStreamEv")
-        if not stream_fn:
-            return 0
-        stream_ptr = ctypes.CFUNCTYPE(ctypes.c_void_p)(stream_fn)()
-
-        cb_fn = dlsym(handle, b"_ZN2at3mps9MPSStream13commandBufferEv")
-        if not cb_fn:
-            return 0
-        cb_ptr = ctypes.CFUNCTYPE(ctypes.c_void_p, ctypes.c_void_p)(cb_fn)(stream_ptr)
-
-        objc = ctypes.CDLL("/usr/lib/libobjc.A.dylib")
-        sel_reg = objc.sel_registerName
-        sel_reg.restype = ctypes.c_void_p
-        sel_reg.argtypes = [ctypes.c_char_p]
-        msg_send = objc.objc_msgSend
-        msg_send.restype = ctypes.c_void_p
-        msg_send.argtypes = [ctypes.c_void_p, ctypes.c_void_p]
-
-        queue_ptr = msg_send(cb_ptr, sel_reg(b"commandQueue"))
-        return queue_ptr or 0
-    except (OSError, AttributeError):
-        return 0
-
-
 ########################## init ##########################
 def init(
     *,
@@ -294,7 +258,7 @@ def init(
     # On Metal, share PyTorch MPS's command queue with Quadrants so that GPU-side ordering is automatic and the
     # per-interop-point sync overhead (qd.sync / torch.mps.synchronize) is eliminated.
     if backend == _gs_backend.metal and device.type == "mps":
-        mps_queue = _get_mps_command_queue()
+        mps_queue = qd.interop.get_mps_command_queue()
         if not mps_queue:
             raise_exception(
                 "Failed to extract PyTorch MPS's Metal command queue. "