Release 2.4.0

[v2.4.0] - 2026-07-21

• Performance — Raising the bar once again: Typical workloads improve by approximately 10%–20%; actual gains vary with thread count, log format, argument types, and output mode.
• Bug fix: Slight memory leak #70 — internal caches of the compressed file Appender are now strictly bounded, completely resolving the issue where memory usage could keep growing in certain cases.
• Bug fix: Fixed parsing of log.buffer_policy_when_full; the configured discard, block, or expand policy is now applied correctly instead of silently falling back to the default block policy.
• Configuration: Added write_cache_size for file Appenders, allowing the per-Appender write cache to be tuned from 64 KiB to 4 MiB.
• Configuration: Added format_template_cache_max_entries and thread_info_cache_max_entries for compressed file Appenders. Defaults remain 100000 and 2048; exceeding the configured entry limit keeps memory bounded but may emit repeated templates and increase compressed file size.
• Robustness: Simplified the compressed-appender lookup cache (removed the redundant hot sub-table) and hardened its slot hashing with a murmur3-style full-avalanche finalizer, so the known structured key families (4-aligned thread ids, high-bits-only entropy, XOR/multiplier-cancelling keys) can no longer collapse the table into a single probe chain.