Add support for GPU build environment

Currently build environment is hardcoded to `LINUX_CONTAINER`. This will cause builds that involve compiling Cuda operations to fail (for example, Docker images that include custom PyTorch or Tensorflow operations). Adding an `--environment` option to change build environment to `LINUX_GPU_CONTAINER` fixes this. This also requires changing the `--compute-type` argument to `BUILD_GENERAL1_LARGE`.