Improve the cache

1) We need to improve the cache; it is not good at the moment;
- Writing and loading a single file is suitable for BNCI but not for large datasets. (implemented at the moment)
- Writing single files for the epoch is far from perfect. (implemented at the moment)

The issue names no files, tests, or entry points, so first locate the cache implementation and clarify the intended behavior for large datasets and epoch files. Define measurable scalability requirements and tests before changing the cache; the work is done when those requirements are met.

Written by the indexing model from the issue text.