graphics/opencv: Update to 3.4.1.

== OpenCV 3.4.1

dnn
- Added support for quantized TensorFlow networks
- OpenCV is now able to use Intel DL inference engine as DNN
  acceleration backend
- Added AVX-512 acceleration to the performance-critical kernels, such
  as convolution and fully-connected layers
- SSD-based models trained and retrained in TensorFlow Object
  Detection API can be easier imported by a single invocation of
  python script making a text graph representation
- Performance of pthreads backend of cv::parallel_for_() has been
  greatly improved on many core machines
- OpenCL backend has been expanded to cover more layers
- Several bugs in various layers have been fixed

OpenCL
- On-disk caching of precompiled OpenCL kernels has been fixed to
  comply with OpenCL standard
- Certain cases with UMat deadlock when copying UMats in different
  threads has been fixed

Android
- Supported Android NDK16
- Added build.gradle into OpenCV 4 Android SDK
- Added initial support of Camera2 API via JavaCamera2View interface

C++
- C++11: added support of multi-dimentional cv::Mat creation via C++
  initializers lists
- C++17: OpenCV source code and tests comply C++17 standard

Misc
- opencv_contrib: added GMS matching
- opencv_contrib: added CSR-DCF tracker
- opencv_contrib: several improvements in OVIS module

== OpenCV 3.4

- New background subtraction algorithms have been integrated.

dnn
- Added faster R-CNN support
- Javascript bindings have been extended to cover DNN module
- DNN has been further accelerated for iGPU using OpenCL

OpenCL
- On-disk caching of precompiled OpenCL kernels has been finally
  implemented
- It's now possible to load and run pre-compiled OpenCL kernels via
  T-API
- Bit-exact 8-bit and 16-bit resize has been implemented

(fhajny)