Import Google's protobuf matchers via Nucleus.

[SanjayVas]

Google's protobuf matchers are open-sourced in the Nucleus project. This imports those protobuf matchers by replacing the dependency on TensorFlow with glog.

See google/googletest#1761

---

This change is 

[stevenwarejones]

Reviewed 7 of 7 files at r1, all commit messages.
Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @SanjayVas)

---

build/nucleus/repo.bzl line 27 at r1 (raw file):

        sha256 = "326907b1ae0a3419806b52ff1bd80502d842248edc7986a89fba9fc4ed37ff0b",
        strip_prefix = "nucleus-0.6.0",
        patches = ["@wfa_common_cpp//build/nucleus:testing.patch"],

isn't this going to be hard to maintain? Not sure what I think about going this route. What's the shortcoming in our current protobuf matchers that we need to use a patched nucleus intead?

[SanjayVas]

Reviewable status: all files reviewed, 1 unresolved discussion (waiting on @SanjayVas)

---

build/nucleus/repo.bzl line 27 at r1 (raw file):

What's the shortcoming in our current protobuf matchers

At a glance:

• Supporting text format protos in addition to message instances.
• Feature parity with ProtoTruth:
  • Ignoring specific fields.
  • Partial matching (similar to ProtoTruth's comparingExpectedFieldsOnly)
  • Approximate matching for floating point values.

isn't this going to be hard to maintain?

Protobuf matchers in Nucleus are have not changed in years. Last update was in 2018.

That said, I think I found a "clean" version in https://github.com/google-research/tf-opt/tree/master/tf_opt/open_source that's both more up-to-date and does not depend on TensorFlow. We may be able to pull that in without patching. Let me investigate.