chore: release v0.6.8

[MagicalTux]

🤖 New release

• compcol: 0.6.7 -> 0.6.8 (✓ API compatible changes)

Changelog

0.6.8 - 2026-07-07

Fixed

• large-file corruption in snappy/lzss/lzs/xpress + stress harness (#116)

Other

• (decode) vectorize match-copy loops (brotli, sit13, rar1/2/5) (#115)
• (zstd,lzfse) vectorize self-overlapping match copy on decode (#114)
• (fuzz) add round-trip fuzz target for the encoders (#113)
• compression optimizations (match extension, huffman decode, xz/lzma2 ratio) (#112)
• (lzss,huffman) hash-chain match finder + table Huffman decode
• (brotli,zstd) faster encode on low-redundancy input

Changed

• (lzss) replaced the encoder's O(N·n) brute-force ring-buffer match scan
  (every position compared against all 4096 ring slots) with a hash-chain
  finder over the raw input, translating each match source to the ring index
  the decoder expects. Encode of low-redundancy input is dramatically faster —
  ~9× on natural-language text and ~700× on incompressible input (which had
  collapsed to ~0.3 MB/s) — with the compressed size unchanged (match lengths,
  which determine output size, are preserved; only the tie-broken source
  position can differ).
• (huffman) the standalone canonical-Huffman decoder now decodes via a single
  peek-and-lookup table (indexed by the next max_length bits) instead of
  walking each code one bit at a time, roughly halving decode instruction count
  (~1.9–2.1× fewer) across text and high-entropy input. Output is unchanged and
  corrupt/truncated streams are still rejected without panicking.

---

This PR was generated with release-plz.