blender: build hip kernels

Without this, things need to be built at runtime. This matches what is
done for CUDA.

Technically we are only using `hipcc` (gotten here via `clr`) at build
time, however it is complicated in that Blender assumes all ROCm things
are together, and so *just* having `hipcc` causes it to get confused a
bit (especially later when we add hiprt support). While setting
`HIP_ROOT_DIR` might help a bit, it’s all a bit fragile and we have
`clr` in the closure already anyway so it doesn’t seem worth it to
over-optimize. Using `clr` brings libraries and such, so add it to
buildInputs instead of nativeBuildInputs to be a bit safe.

pkgs/by-name/bl/blender/package.nix

@@ -194,6 +194,9 @@ stdenv'.mkDerivation (finalAttrs: {
     "-DOPTIX_ROOT_DIR=${optix}"
     "-DWITH_CYCLES_CUDA_BINARIES=ON"
   ]
+  ++ lib.optionals hipSupport [
+    "-DWITH_CYCLES_HIP_BINARIES=ON"
+  ]
   ++ lib.optionals waylandSupport [
     "-DWITH_GHOST_WAYLAND=ON"
     "-DWITH_GHOST_WAYLAND_DBUS=ON"
@@ -268,6 +271,7 @@ stdenv'.mkDerivation (finalAttrs: {
     zstd
   ]
   ++ lib.optional embreeSupport embree
+  ++ lib.optional hipSupport rocmPackages.clr
   ++ lib.optional openImageDenoiseSupport (openimagedenoise.override { inherit cudaSupport tbb; })
   ++ (
     if (!stdenv.hostPlatform.isDarwin) then