Query as Entities

[relationship](https://github.com/james-j-obrien/rfcs/blob/minimal-fragmenting-relationships/rfcs/79-minimal-fragmenting-relationships.md#query-and-system-caches) requires a more performant and powerful way to manage query's state, as archetype fragmentation becomes prevalent when relationships are used.

This goal can be achieved through 'Query as Entities' – a pattern that selectively exposes new/deleted archetypes only to affected queries requiring iteration over them.

There are some excellent PRs that have tried to implement it for Bevy: #14668 , #18860

Implementation Steps:

- [x] Trigger an event when a new archetype is created.
- [ ] Componentize system-internal QueryState.
- [ ] Use Observer to manage query entities.
- [ ] Type-erased QueryState Component.
- [ ] (optional) Provide users with a more ergonomic way to use managed QueryState.
- [ ] (optional) To optimize, only notify observers of impacted query entities instead of iterating over every query entity.

Start with the linked relationship RFC and the earlier Bevy PRs #14668 and #18860 to understand the proposed Query-as-Entities design. Then inspect the existing event for new archetypes and the system-internal QueryState and Observer entry points. Done means the remaining implementation steps are addressed, including componentized and type-erased QueryState and observer-managed query entities.

Written by the indexing model from the issue text.