[CFL] Load luma as prediction for chroma

Loads the stored reconstructed luma pixels for each trasnform block
inside a prediction block. Supports 4:4:4 and 4:2:0 chroma subsampling
modes.

The CFL_CTX struct is now in cfl.h with appropriate forward declarations

Change-Id: I44c117899414a10a8318d14ecaed402f803de97d
6 files changed