fix tensorflow build and mention v1.13.0 in the doc (LeelaChessZero#774)

Author: borg323

src/neural/network_tf.cc

@@ -205,6 +205,7 @@ class TFNetworkComputation : public NetworkComputation {
   float GetQVal(int sample) const override {
     return output_[0].template matrix<float>()(sample, 0);
   }
+  float GetDVal(int sample) const override { return 0.0f; }
   float GetPVal(int sample, int move_id) const override {
     return output_[1].template matrix<float>()(sample, move_id);
   }

tensorflow.md

@@ -1,11 +1,12 @@
 To build with tensorflow under linux you need to install Tensorflow_cc from
-<https://github.com/FloopCZ/tensorflow_cc>. Either release v1.9.0 or v1.12.0.
-Tensorflow_cc requires a specific version of protobuf, which constrains the
-build. Release v1.9.0 works out of the box, since the default protobuf
+<https://github.com/FloopCZ/tensorflow_cc>. Either release v1.9.0, v1.12.0 or
+v1.13.0. Tensorflow_cc requires a specific version of protobuf, which constrains
+the build. Release v1.9.0 works out of the box, since the default protobuf
 subproject (v3.5.1) is compatible and is used instead of a system installed
-version. In contrast release v1.12.0 needs protobuf v3.6.0 which can be built
-by adding `-Dprotobuf-3-6-0=true` to the build command line. Note that this
-protobuf version has issues with static builds and crashes so is not
+version. In contrast release v1.12.0 needs protobuf v3.6.0 and release v1.13.0
+is built with protobuf 3.6.1 but also works with 3.6.0. For those versions
+`-Dprotobuf-3-6-0=true` should be added to the build command line. Note that
+this protobuf version has issues with static builds and crashes so is not
 recommended for normal use. The crashes look very similar to:
 * <https://github.com/protocolbuffers/protobuf/issues/5107>
 * <https://github.com/protocolbuffers/protobuf/issues/5353>