Context:

After r1.2, TensorBoard moved out of the TensorFlow repository, into its own
repository and its own pip package (presently tensorflow-tensorboard, will later
switch to just tensorboard). The new pip package specifies the `tensorboard`
command, so I removed it from the list of console scripts forTensorFlow. I also
added tensorflow-tensorboard as a pip dependency.

However, it turns out that the pip order of operations is:
- install pip dependencies (thus getting tensorflow-tensorboard and the new
  tensorboard command)
- remove deprecated console scripts (thus erroneously removing the new pointer
  to tensorboard)

To fix this, I returned the `tensorboard` console script to tensorflow's
setup.py, except it now references the tensorboard package rather than the
tensorflow package. Thus, the console script declaration in tensorflow and
tensorboard are identical. We can be confident that the tensorboard package is
available, because it is specified by the pip dependency.

Test Plan:

- Create a clean virtualenv.
- pip install tensorflow < 1.3.
- verify that the tensorboard command works properly
- pip install tensorflow 1.3 using a pip package generated with this change
- verify that the tensorboard command still works

It now takes about 400ms rather than 800ms, if the file system cache is warm.

Most of the latency was due to parsing text_format OpList protocol buffers in
our generated sources. We now use a binary representation, while preserving the
text proto as a comment for readability.

Note: This change does not improve the latency of dereferencing tf.contrib,
which takes about 340ms.

Every summary op writes data for a single plugin to process. Hence, each
SummaryMetadata proto should have a single PluginData optional field
(instead of a repeated one). This removes much complexity from
TensorBoard logic that loops over the plugin data. It also simplifies
the SQL schema - it can now enforce a one-to-one relationship between
summary op and plugin.

Make 'import tensorflow' go faster

…to r1.3

Turn off grappler for 1.3

Make plugin_data optional instead of repeated

Updating version to rc2.

Pin 1.3.x package to tensorflow-tensorboard 0.1.x

…rd-install

Ensure that TensorBoard is still available when pip installed. (r1.3)

Make layout optimizer_test manual due to Grappler being off in the r1.3 branch.

In #11952, I had set made some logic within debug_grpc_testlib return too early, breaking some debugger-related behavior. This PR fixes that.

Fix debugger logic in r1.3

Move Estimators to beginning of Prog. Guide.
Change title of Datasets unit.

PiperOrigin-RevId: 167314186

Fix broken link in Estimators.

PiperOrigin-RevId: 168159289

Update docs symposium

* fix broken links, add links check to sanity

* fix broken link in export.md

Updating protobuf and llvm hashes.

TF Docs fix for 1.3

GitHub only 1.3.1 release.

Fixes #13377

Rename set to depset (#13443)

This makes the definition consistent with TensorBoard and TensorFlow
Serving. It's better to track HEAD than the release versions.

PiperOrigin-RevId: 170788851

I find the comment say that embedding shape is [num_encoder_symbols x input_size], I was so confused and checked the source code, I believe it will be better to change the input_size to embedding_size