[Bug] Prompt caching not persisting across server reconnections, causing duplicate token usage

Question

**Bug Description**
I have made a /btw ask and my 5 hour usage jumped by 12 percentage points. Then, my server environment hangs (I am running Claude on a research cluster), and I had to resume the conversation from a different login node. I sent another prompt to the same conversation (this one), and the usage jumped by another 12 percentage points. This looks like a bug in prompt caching -- the cloud is not keeping up with context just sent over and only categorizing caching based on the insta