Improve error message thrown when cached queries are created while world is in readonly mode

[Indra-db]

see related issue in rust:
Indra-db/Flecs-Rust#220

To Reproduce

void bar(flecs::world& world) {
    auto q = world.query_builder<>()
        .with<Foo>(flecs::Wildcard)
        .group_by<Foo>()
        .build();

    q.run([](flecs::iter& it) {});
}

void test_crashd() {
    flecs::world world;

    world.system<>("").run([&](flecs::iter& it) {
        bar(world);
    });

    world.progress();
}

[](fatal: flecs.c: 40045: assert: !(world->flags & EcsWorldReadonly) INTERNAL_ERROR)](fatal: flecs.c: 40045: assert: !(world->flags & EcsWorldReadonly) INTERNAL_ERROR)

[SanderMertens]

This happens because creating a grouped query makes it cached, and cached queries can't be created from systems that are not immediate since they require mutating the world, and the world will be in readonly mode.

This should not be reported as INTERNAL_ERROR though, will take a look at improving that.