Upgrade the cache policy from TinyLFU to W-TinyLFU

[tatsuya6502]

Currently, Moka cache uses TinyLFU as the admission/eviction policies (with a small buffer in front of the LFU filter). While it works well for many workloads like database, search, and analytics, it will not work well for recency-biased workloads, like job queues and event streams.

To provide better hit rate for such workloads, do the followings:

• Upgrade the cache admission/eviction policy from TinyLFU to Window-TinyLFU (W-TinyLFU).
• Implement hill climbing to auto-tune the size of the LRU window for current workload.

For more details, see the "Eviction Policy" chapter of an article about Java Caffeine cache: http://highscalability.com/blog/2019/2/25/design-of-a-modern-cachepart-deux.html

[tatsuya6502]

I started to implement the Window-TinyLFU policy.

• Diff: https://gitlab.com/moka-labs/moka-gh249-support-w-tinylfu/moka/-/compare/main...w-tinylfu
• An updated driver for Caffeine Simulator.
  • https://github.com/moka-rs/caffeine-sim-drivers/pull/2