[Pytorch] example with actual BigTable access

[kboroszko]

Create an example method where you supply project-id, table name and columns and get a Pytorch Tensor with the results.

[kboroszko]

Decisions, decisions...

• In terms of code structure, to make the setup as simple as possible, all c++ extensions are in a separate package called pbt_C. We later call them from python code by importing that extra package. Nvidia used a similar approach: LINK
• Because we want our lib to have no dependencies we compile all of the libraries statically and with fPIC flag
• We also compile with a flag -D_GLIBCXX_USE_CXX11_ABI=0 every c++ dependency because the libraries need the same version of basic_string as pytorch. For some reason pytorch is compiled with this flag set to 0 which means that we either need to recompile all dependencies the same way or we would need to compile pytorch from source. more details

From pytorch doc:

A small note on compilers: Due to ABI versioning issues, the compiler you use to build your C++ extension must be ABI-compatible with the compiler PyTorch was built with.