[relay][frontend] TensorFlow saved model support

[yongwww]

Add TF saved model support in relay tf frontend, infer shape w/o add_shapes=True.
Please help take a look. @srkreddy1238 @yzhliu @zhiics

[zhiics]

Nit: move this code to an example section?

Examples

.. code-block:: python
    parser = TfParser(model_dir)
    graph = parser.parse()

[yongwww]

added, thx. BTW, backslash like "\----------" could help avoid markdown parsing

[yongwww]

could you please help review it when you have a chance? @srkreddy1238

[srkreddy1238]

Look good to me except that the tensorflow_parser is same for both NNVM and relay.
Consider keeping one copy and use across NNVM and relay. Practice so far was to keep under relay and use it in NNVM.

[tqchen]

@yongwww please update per comment. @srkreddy1238 please try to ping the contributor next time in case ppl forget

[tqchen]

ping @yongwww

[yongwww]

@tqchen I am working on it, will update it later today. Sorry I was working on something else and was in TensorFlow dev summit 2019 the past week.

[yongwww]

@srkreddy1238 Just keep one copy of tensorflow_parser in relay, please take another look.

[srkreddy1238]

Thanks @yongwww.
Just resolve the conflicts & we are good to go.

Also suggest to consider adding a test case or (and) a tutorial to demonstrate (& protect) TFParser use case later after this PR.

[yongwww]

@srkreddy1238 Thanks! Sure I'll do that. I am working on another PR to add more ops mapping in tf frontend, such as Gather, Exp, etc.

[yongwww]

@srkreddy1238 conflict solved

[srkreddy1238]

Thanks @yongwww LGTM