Refactoring of HPMV (history based marge candidates), moving the LUTs from Slice to CodingStructure
In order to ever be able to support some kind of sub-slice level parallelization (see #201), the Slice object cannot hold values which change during the encoding of a slice (this is more or less equivalent to using global variables).
In this MR, HPMV LUTs are refactored from Slice to CodingStructure.
I tested 49fr for all but A classes and the results are as expected 0%/0%/0% BD-Y/Cb/Cr 100%/100% EncT/DecT.