UV transform type inference issue when q_idx=0 but not lossless
See !869 (closed). The issue is that for (chroma) intra predictors, the non-lossless-but-inferred-because-qidx=0 (which is basically a bitstream bug) should be from the lookup table. In most cases, the result is still DCT_DCT, but for directional predictors, it can be one of the ADST variants also.