When no heuristics match, still pick a fallback view if possible

[jleibs]

As of #3323 our space view part systems now use indicator components to filter the heuristics. This means if we just log via the new as_components APIs, we don't get any views. Patching in indicators is still a bit painful, so it would be nice if we had sane fallback behavior.

[emilk]

Let's do it for 0.9 if it is easy

[Wumpf]

The basics of implementing this are easy as expected. Instead of discarding entities based on heuristics from the EntitiesPerSystemPerClass list immediately we can do something like this:

pub enum ViewPartHeuristicFilterResult {
    /// The entity path has all required components for the part system and passes the heuristic filter.
    ///
    /// This means that the PartSystem will with very high likelihood display something meaningful for this entity.
    Passes,

    /// The entity path has all *required* components for the part system, but fails the heuristic filter.
    ///
    /// This typically means that it's either lacking an indicator component or
    /// has some property that make not possible to be displayed by the given part system.
    Fails,
}

pub type EntitiesPerSystem =
    IntMap<ViewSystemName, IntMap<EntityPath, ViewPartHeuristicFilterResult>>;

pub type EntitiesPerSystemPerClass = IntMap<SpaceViewClassName, EntitiesPerSystem>;

We then can use the heuristic result to spawn Space Views where the heuristic passes or, if no heuristic passes, spawn the first one that still fulfills the requirements but not the heuristics. (Would be nice to have a better heuristic than "the first")
For this we'd likely forward this new piece of information to SpaceViewClass::auto_spawn_heuristic

The tricky part that requires a bit deeper changes is how to add entities to an existing space view: The problem is that we do this in a localized manner right now. Each Space View looks at the list of entities its systems would be interested in and then decides whether to add it. Previously, this list was already filtered by the heuristic. With the above altered datastructure however we need to make a choice depending on whether another Space View was better suited for an entity that didn't pass those heuristics. This means that the decision which incoming entities to add to which existing Space View becomes a central one and not one we can solve by looking at one (existing) Space View at a time.

➡️ In the near future the heuristic pass needs pass need to be structured like this (not too unsimilar to now):

• compute EntitiesPerSystemPerClass with heuristic information (note that this is affected by the timeline position and incoming data)
• if enabled, add new space views, using an enhanced SpaceViewClass::auto_spawn_heuristic. TODO: Ideally we rewrite this step to not yet populate entities so this can be done in the next step?
• for all entities that aren't in any space view yet at all determine where they should go among the lists of Space Views with the auto add entities flag enabled

The change for the last step is a bit too much and too dangerous for 0.9. Delayed to 0.10

[emilk]

Maybe related:

• rr.Tensor not working for 1D tensors #3709