Multiple models inference about hello_tf_c_api

[Neargye]

Try loading two graphs, and create a separate session for each graph.

[VladislavAD]

It seams to work but I'm not sure is this stable enough for production. I read that session.run() is thread safe, but what if we have multiple sessions? Is it still safe to call inference from different threads or applications? Also another problem is that I can't set allow_growth to sessions, just can't find example how to do it using C API. I set memory limit successfuly with help of #21 , but I don't know how to set right amount for any loaded graph. So I limit GPU memory per graph to 0.3 from all amount and I can load up to 3 models, despite I know that some graphs are smaller and perhaps some graphs can be bigger in future.

[Neargye]

Unfortunately, I have not used more than 1 graph.

[razr191120]

I have Done it multi model inf concurrently on gpu.