Describe the bug

In the ASV benchmark time_read_from_epoch, we preload the cache back to the beginning, but then ask for an as_of time that is earlier than any versions. I think previously given that the cache hasn't expired we would detect that the prior load was a LOAD_ALL and not reload it, because LOAD_ALL trumps everything. However now we're just looking at the earliest loading time, seeing that the requested time is before that and ignoring the fact that we loaded everything recently, so there's no point doing it again.

Steps/Code to Reproduce

Why do I have to fill in all these boxes?

Expected Results

It is less slow

OS, Python Version and ArcticDB Version

more boxes

Backend storage used

Seriously?

Additional Context

No response