Fix wrong size for array of coeff delta idx #798
This bug resulted in out-of-bounds memory access when syntax element 'alf_luma_coeff_delta_idx[filtIdx]' had a value between 25 (MAX_NUM_ALF_CLASSES) and 32 (normally max value).
This max value is explained in 7.4.3.16.
This resulted in read on uninitialized value, that leads vtm to have non deterministic behavior for a given stream (resulting in yuv differences).
Check was added on values that had an issue, and a new constant describes this max value.