feat: implement Best compression level (zstd level 11, btlazy2 strategy)

[polaz]

Summary

CompressionLevel::Best currently triggers unimplemented!(). Level 11 uses binary tree with lazy matching — highest compression ratio in the current enum.

Analysis

C reference implementation (zstd level 11)

• Strategy: ZSTD_btlazy2 (binary tree + lazy2)
• Parameters: windowLog=24, chainLog=25, hashLog=24, searchLog=7, minMatch=3, targetLength=128
• Key file: lib/compress/zstd_lazy.c (ZSTD_insertDUBT1, ZSTD_DUBT_findBestMatch)

Algorithm

1. Binary search tree — sorted by suffix content for O(log n) match finding
2. DUBT (Doubly-Unsorted Binary Tree) — insertion marks entries as sorted/unsorted (ZSTD_DUBT_UNSORTED_MARK = 1)
3. Common-length optimization — commonLengthSmaller/Larger to skip already-compared bytes during tree traversal
4. Lazy2 on top — same 2-step look-ahead as level 7 but with btree match finder instead of hash chains

What needs to be implemented

1. Binary tree data structure — left/right pointers in chain table, sorted insertion
2. DUBT insertion — insertDUBT1 equivalent with sorted/unsorted mark handling
3. Tree search — ZSTD_DUBT_findBestMatch with common-length caching
4. Integration with lazy2 — same lazy evaluation but better matches from btree

Acceptance criteria

• CompressionLevel::Best no longer panics
• Roundtrip + interop tests pass
• Compression ratio within 10% of C zstd level 11 on silesia corpus
• Noticeably better ratio than Better level

Dependencies

• feat: implement Better compression level (zstd level 7, lazy2 strategy) #6 (Better level) — shared lazy matching infrastructure

Time estimate

4d

Blocked by

• feat: implement Better compression level (zstd level 7, lazy2 strategy) #6 (Better level — shared lazy matching infrastructure)