OpenCL backend primitive compiler

[vasucp1207]

This only implements the primitive operations using rust ocl crate. More to come in the future. Related to #146

f16 support is not currently available because the ocl::OclPrm trait does not implement f16.

cx.compile(OpenCLCompiler::<f16>::default(), &mut b);

Maybe we need a custom wrapper for solving this.

[jafioti]

incredible work! i'll take a look and have it merged soon. in a week or so we'll have opencl and vulkan merged