Fix post-GC purging of streams view time histogram

[emilk]

What

• Closes Garbage collection (--memory-limit) not correctly applied to streams histogram #2517

The datastore is not the only place we store data. For each node in the entity tree we store a time histogram of where we have data. This was never properly purged post GC - a very rough heuristic was instead used (throw away everything up to the oldest time in the store - which will be an even worse heursistic after #3357).

With this PR, the store gc will book-keep exactly what was deleted, and it will be properly purged from all secondary indices.

The resulting code is a bit complicated, because the store has no idea about the hierarchical nature of the entity paths, but we store the time histograms (and other book-keeping) for each node. That is, logging to /foo/bar we will note the data on /, /foo and /foo/bar, but the data will only be registered in the store for /foo/bar.

I also fixed a bunch of other smaller things that came up.

Best reviewed commit-by-commit.

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 demo.rerun.io (if applicable)

• PR Build Summary
• Docs preview
• Examples preview
• Recent benchmark results
• Wasm size tracking

[jleibs]

Aside from the bug on the component-book-keeping this looks good. In reality that case is pretty rare, but still seems worth fixing for completeness.

[emilk]

Thanks for that @jleibs, it wasn't too bad keeping track of the individual components, and the code makes much more sense now imho!