Account for chroma subsampling when allocating cbi buffers
Decoding Chimera 1080p 8bpc with --threads 16 --framedelay 4
, before and after:
Type Allocs Reuses Share Peak size
---------------------------------------------------------------------
Block data 8 0 19.2% 21 012 480
---------------------------------------------------------------------
9101 49096 109 569 256
Type Allocs Reuses Share Peak size
---------------------------------------------------------------------
Block data 8 0 17.9% 19 353 600
---------------------------------------------------------------------
9101 49096 107 910 376