Build ignores the environment variable DGLBACKEND=tensorflow and still import torch

## 🐛 Bug

The [documentation](https://docs.dgl.ai/install/index.html#working-with-different-backends) says that DGLBACKEND sets the backend. However, with ```DGLBACKEND=tensorflow``` it still imports torch:

```
make[2]: Leaving directory '/usr/ports/math/dgl/work/dgl-1.1.0/third_party/libxsmm'
make[2]: Nothing to be done for 'libxsmm'.
make[2]: Leaving directory '/usr/ports/math/dgl/work/dgl-1.1.0/third_party/libxsmm'
[100% 2/2] cd /usr/ports/math/dgl/work/dgl-1.1.0/dgl_sparse && /usr/local/bin/cmake -E env CMAKE_COMMAND=/usr/local/bin/cmake CUDA_TOOLKIT_ROOT_DIR= USE_CUDA=OFF BINDIR=/usr/ports/math/dgl/work/.build INCLUDEDIR="/usr/ports/math/dgl/work/dgl-1.1.0/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/dlpack/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/dmlc-core/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/phmap/ /usr/ports/math/dgl/work/dgl-1.1.0/third_party/xbyak/ /usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/include/ /usr/ports/math/dgl/work/dgl-1.1.0/tensoradapter/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/nanoflann/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/libxsmm/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/pcg/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/GKlib /usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/tensorpipe" CFLAGS=-fopenmp=libomp\ -O2\ \ -fPIC\ -O2\ -pipe\ \ -fstack-protector-strong\ -fno-strict-aliasing\ -DUSE_LIBXSMM\ -DDGL_CPU_LLC_SIZE=40000000\ -DIDXTYPEWIDTH=64\ -DREALTYPEWIDTH=32\ -DLINUX\ -D_FILE_OFFSET_BITS=64\ -DNDEBUG\ -DNDEBUG2\ -DHAVE_EXECINFO_H\ -DHAVE_GETLINE\ -O3\ \ -Wall\ -Wl,--exclude-libs,ALL CXXFLAGS=-fopenmp=libomp\ -O2\ \ -fPIC\ -std=c++14\ -O2\ -pipe\ -fstack-protector-strong\ -fno-strict-aliasing\ -DUSE_LIBXSMM\ -DDGL_CPU_LLC_SIZE=40000000\ -DIDXTYPEWIDTH=64\ -DREALTYPEWIDTH=32\ \ -Wall\ -Wl,--exclude-libs,ALL LDFLAGS=-Wl,--warn-common\ \ -fstack-protector-strong bash /usr/ports/math/dgl/work/dgl-1.1.0/dgl_sparse/build.sh python3
FAILED: CMakeFiles/dgl_sparse /usr/ports/math/dgl/work/.build/CMakeFiles/dgl_sparse
cd /usr/ports/math/dgl/work/dgl-1.1.0/dgl_sparse && /usr/local/bin/cmake -E env CMAKE_COMMAND=/usr/local/bin/cmake CUDA_TOOLKIT_ROOT_DIR= USE_CUDA=OFF BINDIR=/usr/ports/math/dgl/work/.build INCLUDEDIR="/usr/ports/math/dgl/work/dgl-1.1.0/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/dlpack/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/dmlc-core/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/phmap/ /usr/ports/math/dgl/work/dgl-1.1.0/third_party/xbyak/ /usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/include/ /usr/ports/math/dgl/work/dgl-1.1.0/tensoradapter/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/nanoflann/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/libxsmm/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/pcg/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/GKlib /usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/include /usr/ports/math/dgl/work/dgl-1.1.0/third_party/tensorpipe" CFLAGS=-fopenmp=libomp\ -O2\ \ -fPIC\ -O2\ -pipe\ \ -fstack-protector-strong\ -fno-strict-aliasing\ -DUSE_LIBXSMM\ -DDGL_CPU_LLC_SIZE=40000000\ -DIDXTYPEWIDTH=64\ -DREALTYPEWIDTH=32\ -DLINUX\ -D_FILE_OFFSET_BITS=64\ -DNDEBUG\ -DNDEBUG2\ -DHAVE_EXECINFO_H\ -DHAVE_GETLINE\ -O3\ \ -Wall\ -Wl,--exclude-libs,ALL CXXFLAGS=-fopenmp=libomp\ -O2\ \ -fPIC\ -std=c++14\ -O2\ -pipe\ -fstack-protector-strong\ -fno-strict-aliasing\ -DUSE_LIBXSMM\ -DDGL_CPU_LLC_SIZE=40000000\ -DIDXTYPEWIDTH=64\ -DREALTYPEWIDTH=32\ \ -Wall\ -Wl,--exclude-libs,ALL LDFLAGS=-Wl,--warn-common\ \ -fstack-protector-strong bash /usr/ports/math/dgl/work/dgl-1.1.0/dgl_sparse/build.sh python3
-DCUDA_TOOLKIT_ROOT_DIR= -DTORCH_CUDA_ARCH_LIST= -DUSE_CUDA=OFF -DDGL_INCLUDE_DIRS=/usr/ports/math/dgl/work/dgl-1.1.0/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/dlpack/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/dmlc-core/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/phmap/;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/xbyak/;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/include/;/usr/ports/math/dgl/work/dgl-1.1.0/tensoradapter/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/nanoflann/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/libxsmm/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/pcg/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/GKlib;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/METIS/include;/usr/ports/math/dgl/work/dgl-1.1.0/third_party/tensorpipe -DDGL_BUILD_DIR=/usr/ports/math/dgl/work/.build
Traceback (most recent call last):
File "", line 1, in
ModuleNotFoundError: No module named 'torch'
ninja: build stopped: subcommand failed.
===> Compilation failed unexpectedly.
```

## To Reproduce

cmake configure + build.

## Environment

- DGL Version (e.g., 1.0): 1.1.0
- Backend Library & Version (e.g., PyTorch 0.4.1, MXNet/Gluon 1.3): TensorFlow
- OS (e.g., Linux): FreeBSD 13.2
- How you installed DGL (`conda`, `pip`, source): build as a FreeBSD port
- Python version: 3.9