Add --fat_apk_cpu=arm64-v8a for arm64 Android builds

Bazel silently uses the wrong build settings for --config=android_arm64 (--cpu=arm64-v8a is not enough), and actually still uses armeabi-v7a in at least some cases. --fat_apk_cpu fixes this.

See tensorflow#15581 for more.

PiperOrigin-RevId: 181525260

tools/bazel.rc

@@ -1,10 +1,14 @@
-# Android configs
+# Android configs. Bazel needs to have --cpu and --fat_apk_cpu both set to the
+# target CPU to build transient dependencies correctly. See
+# https://docs.bazel.build/versions/master/user-manual.html#flag--fat_apk_cpu
 build:android --crosstool_top=//external:android/crosstool
 build:android --host_crosstool_top=@bazel_tools//tools/cpp:toolchain
 build:android_arm --config=android
 build:android_arm --cpu=armeabi-v7a
+build:android_arm --fat_apk_cpu=armeabi-v7a
 build:android_arm64 --config=android
 build:android_arm64 --cpu=arm64-v8a
+build:android_arm64 --fat_apk_cpu=arm64-v8a
 
 # Config to use a mostly-static build and disable modular op registration
 # support (this will revert to loading TensorFlow with RTLD_GLOBAL in Python).