CL: Allow tweaking the local size at runtime
This new version explores the approach proposed by @kusma, that is, delay the the spirv-to-dxil step until we have all information so we don't have to patch the binary, and it's indeed much cleaner :-).
I intend to base local mem size passing on the same approach, let me know if you think that's not appropriate.
Edited by Boris Brezillon