nx_zstd_block.nx

nx_zstd_block.nx -- Zstandard compressed-block decompression. The integration layer: literals section + sequences section -> plain bytes. This is what turns the six proven zstd layers into a decoder that actually handles a Compressed block instead of refusing it. WHAT IS WIRED. Raw, RLE and Huffman-coded literals (with a directly-encoded weight table), plus the zero-sequence case where the block's output IS its literals. That is a complete, legal zstd block -- it is what an encoder emits for data it could not match, which is common for already-compressed or high-entropy input. WHAT IS NOT, AND SAYS SO. FSE-coded sequences and FSE-compressed Huffman weights each return their own distinct refusal code. A decoder that quietly emitted the literals for a block that also carried sequences would produce output that is the right LENGTH and the wrong BYTES -- the worst possible failure, because a length check would pass it. THE HUFFMAN TREE DESCRIPTION IS INSIDE THE COMPRESSED SIZE. compressed_size from the literals header covers the tree description AND the stream(s), so the stream length is compressed_size minus the description. Treating compressed_size as the stream alone over-reads by the description length and desynchronises the sequences section that follows. genealogy_id: zstandard_rfc8878_block lineage_id: nx_zstd_block_v1 license_tier: ORIGINAL