Speed up cosmetic filtering

[atuchin-m]

The PR fixes cosmetic filtering performance that regressed after refactoring cosmetic filters to flatbuffers (binary search is slower than hash set/map)
That's a dealbreaker for rolling 0.11.x.

The PR:

• Introduce new structures FlatHash{Set,Map} that uses open hashing approach. That provide a good balance between performance and memory usage.
• Hasher is used rustc-hash (the same version used in Chromium)
• Add new traits to reuse the current code.
• Memory usage is increased (for 0.11.x: 10 MB => 10.5MB, looks acceptable)

TODO: add unit tests (if the rest looks fine). Done

[github-actions]

Rust Benchmark

Benchmark suite	Current: 388c759	Previous: 1bfadd6	Ratio
rule-match-browserlike/brave-list	2236581787 ns/iter (± 11322539)	2157327098 ns/iter (± 15833464)	1.04
rule-match-first-request/brave-list	1133621 ns/iter (± 8618)	1067208 ns/iter (± 8765)	1.06
blocker_new/brave-list	152545779 ns/iter (± 632857)	175128732 ns/iter (± 2762769)	0.87
blocker_new/brave-list-deserialize	21994108 ns/iter (± 64316)	73769476 ns/iter (± 854420)	0.30
memory-usage/brave-list-initial	10508747 ns/iter (± 3)	18344503 ns/iter (± 3)	0.57
memory-usage/brave-list-initial/max	64817658 ns/iter (± 3)	66961277 ns/iter (± 3)	0.97
memory-usage/brave-list-initial/alloc-count	1534842 ns/iter (± 3)	1616082 ns/iter (± 3)	0.95
memory-usage/brave-list-1000-requests	2516487 ns/iter (± 3)	2551890 ns/iter (± 3)	0.99
memory-usage/brave-list-1000-requests/alloc-count	66788 ns/iter (± 3)	68816 ns/iter (± 3)	0.97
url_cosmetic_resources/brave-list	202280 ns/iter (± 2208)	189708 ns/iter (± 2713)	1.07
cosmetic-class-id-match/brave-list	3438051 ns/iter (± 949610)	5166097 ns/iter (± 1523809)	0.67

This comment was automatically generated by workflow using github-action-benchmark.

[Copilot]

Pull Request Overview

This PR addresses a performance regression in cosmetic filtering that occurred after refactoring to use flatbuffers. The change replaces binary search-based lookups with hash-based lookups to restore performance while maintaining acceptable memory usage (increasing from 10 MB to 10.5 MB).

Key Changes:

• Introduces FlatHashSet and FlatHashMap structures using open hashing with rustc-hash (FxHasher) for improved lookup performance
• Replaces binary search containers with hash-based containers for cosmetic filter rules
• Updates data format version and serialization hash expectations
• Adds comprehensive unit tests for the new hash-based containers

Reviewed Changes

Copilot reviewed 17 out of 18 changed files in this pull request and generated 2 comments.

Show a summary per file

File	Description
src/flatbuffers/containers/hash_index.rs	Core hash table implementation with open addressing and 25-50% load factor
src/flatbuffers/containers/hash_set.rs	HashSet wrapper around HashIndex for set operations
src/flatbuffers/containers/hash_map.rs	HashMap wrapper around HashIndex for key-value storage
src/flatbuffers/containers/fb_index.rs	New trait FbIndex for unified indexed data access
src/flatbuffers/containers/sorted_index.rs	Refactored to use FbIndex trait instead of duplicating methods
src/cosmetic_filter_cache_builder.rs	Updated to use hash-based containers and StringVector wrapper
src/cosmetic_filter_cache.rs	Updated to use HashSetView and HashMapStringView for lookups
src/flatbuffers/fb_network_filter.fbs	Added StringVector table definition
src/flatbuffers/fb_network_filter_generated.rs	Generated code for StringVector support
src/data_format/mod.rs	Renamed constant for clarity (ADBLOCK_FLATBUFFER_VERSION → ADBLOCK_RUST_DAT_VERSION)
tests/unit/engine.rs	Updated expected serialization hashes for new format
tests/unit/flatbuffers/containers/hash_set.rs	Unit tests for HashSet functionality
tests/unit/flatbuffers/containers/hash_map.rs	Unit tests for HashMap functionality
Cargo.toml	Added rustc-hash dependency

---

_{Tip: Customize your code reviews with copilot-instructions.md. Create the file or learn how to get started.}

[antonok-edm]

just the version documentation I mentioned above