Add a local cache

Every read incurs S3 API call. Add a local cache to read the objects.

the cache should be configurable: {in-memory, on-disk, both(tiered)}, size, LRU / Sieve

Start by locating the code path that reads objects from S3 and identify whether reads are centralized. Define the cache configuration and behavior for in-memory, on-disk, and tiered modes, including size and LRU or Sieve eviction. Done means repeated reads avoid unnecessary S3 calls while each supported configuration behaves as specified.

Written by the indexing model from the issue text.