Better change detection benchmarking

[alice-i-cecile]

What problem does this solve or what need does it fill?

The change detection benchmarks we currently have (#4883) don't accurately cover all of the realistic use cases we care about.\

What solution would you like?

We should cover:

1. No entity has changed.
2. Only one entity has changed across many entities.
3. Only entities within a single archetype has changed, across many archetypes and entities.
4. Half of the entities have changed, completely at random.
5. Almost all entities have changed.
6. Almost all of the archetypes have changed.
7. All entities have changed.

We should independently measure both the speed of making the mutations, and the speed of detecting the changes.

Additional context

#11173 improves the reliability of the existing benchmarks, and should be merged first.

This is needed by #4882 and #5097, to evaluate the impacts of changes made to address them.