Fault-injection patch for deterministic reproduction on Windows ASAN.

This does NOT bypass any security check. It simulates the transient
`Direct3D11CaptureFramePool::Recreate()` failure that the report claims is
reachable under GPU memory pressure, so the partial-resize desync (and the
resulting OOB read past the mapped staging texture) fires on every
shrink->restore cycle instead of probabilistically.

Apply, build `chrome` for Windows with is_asan=true, then run poc.html and
perform one shrink->restore. ASAN should report a read past the mapped
D3D11 staging allocation inside the memcpy loop at
wgc_capture_session.cc:720.

--- a/third_party/webrtc/modules/desktop_capture/win/wgc_capture_session.cc
+++ b/third_party/webrtc/modules/desktop_capture/win/wgc_capture_session.cc
@@ -595,9 +595,16 @@ HRESULT WgcCaptureSession::ProcessFrame() {
       return hr;
     }

-    hr = frame_pool_->Recreate(direct3d_device_.Get(), kPixelFormat,
-                               num_buffers(), new_size);
+    // FAULT INJECTION: simulate Recreate() failing on shrink so that
+    // mapped_texture_ (already replaced above) is smaller than size_.
+    if (new_size.Width < size_.Width || new_size.Height < size_.Height) {
+      hr = E_OUTOFMEMORY;
+    } else {
+      hr = frame_pool_->Recreate(direct3d_device_.Get(), kPixelFormat,
+                                 num_buffers(), new_size);
+    }
     if (FAILED(hr)) {
       RecordGetFrameResult(GetFrameResult::kRecreateFramePoolFailed);
       return hr;
     }
   }