Resolves #274 compression filtering

Changes the approach to Chroma upsampling to reduce the chroma blur problem. Also performs a bilinear interpolation of chroma values to smooth the depth. NvPipe has been merged into the repo but also removed from the decoder to rely instead on the latest Nvidia Video SDK. The same must be done for the encode step and then more refinements to compression can be made.

components/codecs/include/ftl/codecs/depth_convert_cuda.hpp

@@ -10,9 +10,16 @@ void depth_to_vuya(const cv::cuda::PtrStepSz<float> &depth, const cv::cuda::PtrS
 
 void vuya_to_depth(const cv::cuda::PtrStepSz<float> &depth, const cv::cuda::PtrStepSz<ushort4> &rgba, float maxdepth, cv::cuda::Stream &stream);
 
+void vuya_to_depth(const cv::cuda::PtrStepSz<float> &depth, const cv::cuda::PtrStepSz<ushort> &luminance, const cv::cuda::PtrStepSz<ushort> &chroma, float maxdepth, cv::cuda::Stream &stream);
+
 void smooth_y(const cv::cuda::PtrStepSz<ushort4> &rgba, cv::cuda::Stream &stream);
 
+void nv12_to_float(const uint8_t* src, uint32_t srcPitch, float* dst, uint32_t dstPitch, uint32_t width, uint32_t height, cudaStream_t s);
+
 }
 }
 
+template <class COLOR32>
+void Nv12ToColor32(uint8_t *dpNv12, int nNv12Pitch, uint8_t *dpBgra, int nBgraPitch, int nWidth, int nHeight, int iMatrix = 0, cudaStream_t s=0);
+
 #endif  // _FTL_CODECS_DEPTH_CONVERT_HPP_