vae tiling improvements: encoding support and adaptative overlap #658

Job	Run time
ubuntu-latest-cmake	1m 24s
macOS-latest-cmake	1m 30s
windows-latest-cmake (noavx, -DGGML_AVX=OFF -DGGML_AVX2=OFF -DGGML_FMA=OFF -DSD_BUILD_SHARED_LIBS...	1m 33s
windows-latest-cmake (avx2, -DGGML_AVX2=ON -DSD_BUILD_SHARED_LIBS=ON)	1m 28s
windows-latest-cmake (avx, -DGGML_AVX2=OFF -DSD_BUILD_SHARED_LIBS=ON)	1m 29s
windows-latest-cmake (avx512, -DGGML_AVX512=ON -DSD_BUILD_SHARED_LIBS=ON)	1m 59s
windows-latest-cmake (cuda12, -DSD_CUBLAS=ON -DSD_BUILD_SHARED_LIBS=ON)	7m 49s
windows-latest-cmake (rocm5.5, -G Ninja -DCMAKE_C_COMPILER=clang -DCMAKE_CXX_COMPILER=clang++ -DS...	7m 45s
windows-latest-cmake (vulkan, -DSD_VULKAN=ON -DSD_BUILD_SHARED_LIBS=ON)	2m 55s
release	0s
	27m 52s

Provide feedback