Lints and dogfood

[therealfrauholle]

Fixes all lint's (cargo clippy --features eager,experimental,ndarray). There is also a little bit of dogfood. Follow commit messages for more context.

A cool library, thanks for the work! While getting to know the code by working with example/xor.rs I was figuring I can contribute a little bit.

I have a few more ideas, maybe I find time to work with those:

• Use Bindgen to generate raw bindings. Bindgen now supports library loading (rust-lang/rust-bindgen@fc5fa9a), so bindings in sys and sys-runtime can be unified. Would fix Runtime linkin - eager support #430. I think it is not necessary to pull the whole tensorflow source, only copy the needed header files.
• Some work can be done on incremental compilation of the tensorflow C library. When no .pc file is available, tensorflow-sys (build.rs) took long in the build process. I manually installed a .pc file. Maybe the env variable CARGO_TARGET_DIR can be used to persistently store build artifacts across incremental builds instead of using OUT_DIR.
• Re-enable the pipeline (lints, doctests). @adamcrume apparently had disk space problems (f77f2d9), maybe that was due to tensorflow-sys (build.rs)?
• Compilation is very slow. It might be due to low file/crate modularity. It might be possible to (just as for the C header files) check in the tensorflow/core/ops/ops.pbtxt and run op-codegen in a build.rs - I never worked with protobuf.