Disallow requesting write resource access in Queries

[vil-mo]

Related to #16843

Since WorldQuery::Fetch is Clone, it can't store mutable references to resources, so it doesn't make sense to mutably access resources. In that sense, it is hard to find usecases of mutably accessing resources and to clearly define, what mutably accessing resources would mean, so it's been decided to disallow write resource access.
Also changed documentation of safety requirements of WorldQuery::init_fetch and WorldQuery::fetch to clearly state to the caller, what safety invariants they need to uphold.

[LikeLakers2]

Suggested change

	/// Readonly accesses resourses registered in [`update_component_access`].
	/// Readonly accesses resources registered in [`update_component_access`].

[chescock]

Yeah, I think this is the right move. I only added the code for mutable resources because it looked simple and not because we had a real use case, and it's clear that I was wrong about it being simple.

We should also remove the has_resource_write() check in AssetChanged. It's the only first-party use of resource access in queries, so users might use it as an example, and leaving that check there might trick them into thinking that mutable resource access is supported.

[chescock]

It might be worth adding a check like

debug_assert!(!state.component_access.access().has_any_resource_write());

so that anyone who tries it gets an error.

[vil-mo]

We should also remove the has_resource_write() check in AssetChanged. It's the only first-party use of resource access in queries, so users might use it as an example, and leaving that check there might trick them into thinking that mutable resource access is supported.

It should not be removed, because user could request write access to the resource in the SystemParam prior to the query

[chescock]

It should not be removed, because user could request write access to the resource in the SystemParam prior to the query

I believe that's checked in <Query as SystemParam>::init_state() after the whole query is built. The check in <AssetChanged as WorldQuery>::update_component_access() only has the FilteredAccess from the current query, not the FilteredAccessSet that includes other system params' access.

[vil-mo]

I believe that's checked in <Query as SystemParam>::init_state() after the whole query is built

It's not checked there

fn init_state(world: &mut World, system_meta: &mut SystemMeta) -> Self::State {
    let state = QueryState::new_with_access(world, &mut system_meta.archetype_component_access);
    init_query_param(world, system_meta, &state);
    state
}

[chescock]

The check is in init_query_param. It's in a method so it can be shared by QueryParamBuilder.

[alice-i-cecile]

Note that write access to resources during queries may be useful for certain index updating strategies. That said, it doesn't seem to actually work, so we should clean this up for now and re-add it later when we have clearer requirements.