Caches per recording

[teh-cmc]

Moves Caches from AppState to the StoreHub, and make them per-recording instead of viewer-wide in the process.

This is:

• a pre-requisite to fix the repeated static logging issues (static=True image logging still stores all the data #7404)
• a necessary step towards proper secondary caching
• a fix for a very old issue: cache memory is not reclaimed when closing a recording

I'm not a fan of the implementation: it's brittle, confusing, and I don't like it. Still, it's a net improvement over the status quo, and fixes a very annoying "leak" for end users.

Before:

24-09-25_17.29.57.patched.mp4

After:

24-09-25_17.29.08.patched.mp4

---

• Part of static=True image logging still stores all the data #7404

Checklist

• I have read and agree to Contributor Guide and the Code of Conduct
• I've included a screenshot or gif (if applicable)
• I have tested the web demo (if applicable):
  • Using examples from latest main build: rerun.io/viewer
  • Using full set of examples from nightly build: rerun.io/viewer
• The PR title and labels are set such as to maximize their usefulness for the next release's CHANGELOG
• If applicable, add a new check to the release checklist!
• If have noted any breaking changes to the log API in CHANGELOG.md and the migration guide

• PR Build Summary
• Recent benchmark results
• Wasm size tracking

To run all checks from main, comment on the PR with @rerun-bot full-check.

[emilk]

Makes sense!

[Wumpf]

Should also fix

• GPU memory not freed when closing recording, but keeping viewer open #6755

but should probably check first

[teh-cmc]

Should also fix

* [GPU memory not freed when closing recording, but keeping viewer open #6755](https://github.com/rerun-io/rerun/issues/6755)

but should probably check first

I checked and it's definitely not the case: Counted GPU barely budges when closing the recording. 😞

[teh-cmc]

(Although maybe the problem is not the GPU memory itself, but the reporting done by Counted GPU, e.g. stats don't get properly updated (or downdated, rather) ? I don't know how Counted GPU is fed.)

[Wumpf]

Counted GPU is fed by what re_renderer holds on to, so yeah something goes sideways there.
I think the caches are not properly cleaning up - the "mesh manager" on re_renderer is probably in a sorry state too