Consider introducing an item data cache

[veloman-yunkan]

The idea first surfaced in #956 (comment)

I propose to think about possible benefits of introducing an item data cache. In some scenarios, clusters may be too large a unit of caching. Consider a usage pattern when multiple relatively small items all from different clusters are constantly used. The total memory consumed by those items can be orders of magnitude smaller than the size of their clusters. And if the latter exceeds the cluster cache limit, libzim may keep thrashing for no good reason.

[kelson42]

@veloman-yunkan Very open about that, but this rease a lot of question to me, like for example:

• Should we consider that the item cache memory comes from the same memory pool like the cluster cache?
• What would be default values for the cache size
• how do both caches (cluster and item) are articulated between each other.