Improve compression support

* [ ] Checksum support (<- Great first issue!)
* [ ] Check if/how much building custom FSE tables improves the compression rate and what the performance implications are
* [ ] Check if/how much rebuilding huffman tables for each block helps/hurts performance and ratio
* [ ] Dictionary support
* [ ] Check other matching algorithms and window sizes to implement more compression levels
* [x] Make the `Matcher` trait public and allow users to provide their own implementations

The issue lists checksum support, custom FSE and Huffman table experiments, dictionary support, matching algorithms, window sizes, and the public Matcher trait, but names no files or tests. Start by narrowing the checklist to one item, then locate the related compression entry points and establish how compression ratio and performance are measured. Done means one scoped improvement with documented results and coverage.

Written by the indexing model from the issue text.