Invaild option for build_pip_package

[petronny]

In build(),

    bazel-bin/tensorflow/tools/pip_package/build_pip_package --gpu "${srcdir}"/tmprocm

bazel-bin/tensorflow/tools/pip_package/build_pip_package is pointed to tensorflow/tools/pip_package/build_pip_package.sh.

However, build_pip_package.sh --help shows no --gpu option in its usage.

Usage:
./build_pip_package.sh [--src srcdir] [--dst dstdir] [options]
./build_pip_package.sh dstdir [options]

    --src                 prepare sources in srcdir
                              will use temporary dir if not specified

    --dst                 build wheel in dstdir
                              if dstdir is not set do not build, only prepare sources

  Options:
    --project_name <name> set project name to name
    --cpu                 build tensorflow_cpu
    --gpudirect           build tensorflow_gpudirect
    --rocm                build tensorflow_rocm
    --nightly_flag        build tensorflow nightly

Maybe it should be changed to --gpudirect or --rocm now.

And please also consider setting the srcdir for build_pip_package, or else it will use mktemp -d -t tmp.XXXXXXXXXX and break the build when /tmp is not big enough.
I guess "${srcdir}"/tmprocm is for dstdir. So the updated command may be like

bazel-bin/tensorflow/tools/pip_package/build_pip_package --src "${srcdir}"/tmprocm-src --dst "${srcdir}"/tmprocm --rocm

[petronny]

And tensorflow-rocm and tensorflow-rocm-opt cannot use the same srcdir. So the other one could be tmpoptrocm-src.

[acxz]

patched with bfe29ff

leaving this patched until we don't see the --gpu option in the [community] package for tensorflow. https://gitlab.archlinux.org/archlinux/packaging/packages/tensorflow/-/blob/main/PKGBUILD?ref_type=heads#L174

[acxz]

Removing the patch label as the upstream PKGBUILD has removed the usage of build_pip_package and uses wheel instead. As such I have reverted the patch. Of course, we will need to add in this patch (at least for building rocm and optrocm in two separate dirs) later on, but I want to sort out the other issues that we encounter in the build first.