Compilation times

[hlky]

Compilation times, especially on Windows, can be very long.

Due to limitations mentioned in #1 we currently need to release multiple modules supporting different shapes.

Codegen

Each use of an operator generates another source file. In many cases the contents of source files for the same operator is near identical, the kernel itself is repeated across each file and the only difference, if any, is the launcher.

Reducing the number of files and duplicated code should improve compilation times.

Windows/CMake/msbuild

An investigation into Windows build system is required. We can potentially use a different cmake generator - Ninja, MinGW etc