Explore caching for the persistent forest backend

### What should be done?

Currently the only cache for the persistent forest backend is the LRU cache built into RocksDB. We should explore caching some subset of trees entirely in memory based on tuneable residency thresholds to improve query times for frequently-accessed trees.

### How should it be done?

Explore it and potentially finalise an implementation if it turns out to be useful.

### When is this task done?

If we have achieved consistent and measurable speedups on representative workflows for the persistent forest.

### Additional context

From this [comment](https://github.com/0xMiden/crypto/pull/876#pullrequestreview-3927322779) on 0xMiden/crypto#876.

The issue references the persistent forest backend and RocksDB's LRU cache. Start by examining the code for the persistent forest backend to understand its structure and how it interacts with RocksDB. Look for existing caching mechanisms and performance benchmarks. The goal is to design and test an in-memory caching layer for frequently accessed trees, measuring speedups on representative workflows.

由索引模型根据 Issue 内容生成。