Enable the full stack of CUDA-related filters via ----enable-cuda-llvm

As at the moment, this FFmpeg build's filtering capabilities are extremely limited in NVENC-based workloads, lacking multiple CUDA-based filters needed for a fully functional GPU-bound transcode.

This patch adds the ``--enable-cuda-llvm` `./configure` parameter to FFmpeg's builds, with the addition of `clang` (and its' dependencies) as build requirements for these extra filters.

The following filters are now additionally available, and without invoking the installation of the proprietary NVIDIA CUDA SDK: 

```sh
ffmpeg -filters | grep cuda
 ... bilateral_cuda    V->V       (null)
 T.. bwdif_cuda        V->V       (null)
 ... chromakey_cuda    V->V       (null)
 ... colorspace_cuda   V->V       (null)
 ... hwupload_cuda     V->V       (null)
 ... overlay_cuda      VV->V      (null)
 ... scale_cuda        V->V       (null)
 ... thumbnail_cuda    V->V       (null)
 T.. yadif_cuda        V->V       (null)
``` 

Licensing, therefore, remains unaffected.

Dockerfile

@@ -59,6 +59,11 @@ RUN \
     automake \
     bzip2 \
     cmake \
+    clang \
+    clang-tools \
+    libclang1 \
+    clang-format \
+    clangd \
     diffutils \
     doxygen \
     g++ \
@@ -76,6 +81,13 @@ RUN \
     libtool \
     libv4l-dev \
     libwayland-dev \
+    libllvm-ocaml-dev \
+    libllvm12 \
+    llvm \
+    llvm-dev \
+    llvm-runtime \
+    lldb \
+    lld \
     libx11-dev \
     libx11-xcb-dev \
     libxcb-dri3-dev \
@@ -763,6 +775,7 @@ RUN \
     --enable-nonfree \
     --enable-nvdec \
     --enable-nvenc \
+    --enable-cuda-llvm \
     --enable-opencl \
     --enable-openssl \
     --enable-stripping \