Avijit/fix license files

[avijit-nervana]

We discovered that the license file names are wrong when nGraph build is selected. This PR fixes these file names so that nGraph is built with TensorFlow when requested. This PR also fixes nGraph unit tests due to recent changes in the MKL bazel files.

[aaroey]

What's the actual difference after applying clean_dep?

[avijit-nervana]

We have a C++ unit test (google test based) that depends on TensorFlow C++ as well as core.
Here's the target dependencies defined in third_party/ngraph/ngraph_tf.BUILD:

tf_cc_test(
    name = "ngraph_tf_tests",
    size = "small",
    srcs = [
        "test/tf_exec.cpp",
        "test/main.cpp",
    ],
    deps = [
        ":ngraph_tf",
        "@com_google_googletest//:gtest",
        "@org_tensorflow//tensorflow/cc:cc_ops",
        "@org_tensorflow//tensorflow/cc:client_session",
        "@org_tensorflow//tensorflow/core:tensorflow",
...

One or more of the TF dependencies are defined as tf_cc_binary(...) and the definition of tf_cc_binary(...) includes the conditional dependencies of MKL as shown in line 399 above.
It turns out that without the clean_dep(...) or str(Label(...)) which is equivalent, this dependency is resolved as @ngraph_tf//third_party/mkl which is causing the bazel error such as the following:

ERROR: /tmp/avijitch/.cache/_bazel_avijitch/7ac9ae669f8ee6504932fa4d1b9f916b/external/ngraph_tf/BUILD.bazel:67:1: no such package '@ngraph_tf//third_party/mkl': BUILD file not found on package path and referenced by '@ngraph_tf//:ngraph_tf_tests'

After making this changes, the dependencies are properly expressed and the bazel errors are gone as it's now being correctly resolved as @org_tensorflow//third_party/mkl:BUILD.

Our observation is that the ngraph unit test build started to fail since the MKL bazel files changes went in a couple of weeks ago. Comparing the before and after those changes, we noticed that several bazel files properly used the str(Label(...)) but some didn't (like the ones I added in this PR).

After making the changes proposed in the PR, the ngraph test build and runs.

Please let me know if you need more details or have any questions. Thanks :)

BTW - I used the following bazel query which was very helpful debugging this:
bazel query 'buildfiles(deps(@ngraph_tf//:ngraph_tf_tests))' --keep_going

[aaroey]

Thanks so much for the detailed explanation! LGTM.