Lossless deblock issues

See !875 (merged)

There are basically two issues. First, it is indeed true that deblock should be disabled in lossless coding, but that is not always true, and this is about the normative implications of these special-case situations.

The first bug is that in coded (non-skip) lossless blocks, particularly large ones, we need to deblock all edges. However, our current lf_mask.c implementation only deblocks a few levels deep as if it were non-lossless.

The second bug is that contextually, for lossless skip blocks, the block-edge "tx" value (implying filter_length) should be setto TX_4X4, not the block size or max tx-sz for this block if it were non-lossless.