Updating cache for query with `up` traversal fails when the matched component is removed

[indianakernick]

Describe the bug

An internal error is triggered after a component is removed from an entity that a cached query previously found via up traversal.

To Reproduce

int main() {
    struct Tag {};

    flecs::world world;

    world.component<Tag>();
    world.system().with<Tag>().up().each([](flecs::entity e) {
        std::cout << e << '\n';
    });
    world.system().run([&](flecs::iter &) {
        world.remove_all<Tag>();
    });

    flecs::entity parent = world.entity();
    parent.child();

    world.progress(); // nothing happens
    parent.add<Tag>();
    world.progress(); // system matches child entity
    world.progress(); // crashes
}

The error is this one:

fatal: cache_iter.c: 42: assert: tr != NULL INTERNAL_ERROR

Disabling caching on the system with .cache_kind(flecs::QueryCacheNone) is a workaround.

Expected behavior

This should not crash.

Additional context

Flecs version 4.1.0.