Use weaker version of ComponentDescriptor for indicators (#9990)

### Related

* Part of #6889.
* From
#9938 (comment):

> was curious and poked a lil bit into the ci failure: took me a while
to notice but the primary issue presented by the failing images is that
it draws lines instead of points.
> 
> Which implies an issue with indicators (wtf why is that the only test
failing then.. but sure usually there's no choice; have only a vague
hunch about that :/). Looking at `impl<A: Archetype>
crate::ComponentBatch for GenericIndicatorComponentArray<A>` indicators
are never logged with their full descriptor, but are advertised as such.
> 
> So I quickly hacked the generated `SeriesPoints::descriptor_indicator`
to be name only and sure enough that fixed the test. What to do about
this knowledge? Uhm.. yeah... that is definitely a Monday problem 😄

### What

Because indicators are about to go away soon (™️) anyways, making them
weaker (qualification-wise) is probably the easiest choice.

Doing this in a separate PR to help with potential `git bisect` in the
future.

Author: grtlr

crates/build/re_types_builder/src/codegen/rust/api.rs

@@ -1131,7 +1131,6 @@ fn quote_trait_impls_for_archetype(reporter: &Reporter, obj: &Object) -> TokenSt
             }
         })
         .chain(std::iter::once({
-            let archetype_name = &obj.fqname;
             let indicator_component_name = format!(
                 "{}Indicator",
                 obj.fqname.replace("archetypes", "components")
@@ -1144,7 +1143,7 @@ fn quote_trait_impls_for_archetype(reporter: &Reporter, obj: &Object) -> TokenSt
                 #[inline]
                 pub fn descriptor_indicator() -> ComponentDescriptor {
                     ComponentDescriptor {
-                        archetype_name: Some(#archetype_name.into()),
+                        archetype_name: None,
                         component_name: #indicator_component_name.into(),
                         archetype_field_name: None,
                     }

crates/store/re_chunk/src/migration.rs

@@ -9,6 +9,29 @@ use re_types_core::arrow_helpers::as_array_ref;
 use crate::Chunk;
 
 impl Chunk {
+    /// We look for indicator component descriptors that have an archetype name and strip it.
+    ///
+    /// It turns out that too narrow indicator descriptors cause problems while querying.
+    /// More information: <https://github.com/rerun-io/rerun/pull/9938#issuecomment-2888808593>
+    #[inline]
+    pub fn patched_weak_indicator_descriptor_023_compat(&self) -> Self {
+        let mut chunk = self.clone();
+
+        chunk.components = chunk
+            .components
+            .0
+            .drain()
+            .map(|(mut descriptor, list_array)| {
+                if descriptor.component_name.is_indicator_component() {
+                    descriptor.archetype_name = None;
+                }
+                (descriptor, list_array)
+            })
+            .collect();
+
+        chunk
+    }
+
     /// A temporary migration kernel for blueprint data.
     ///
     /// Deals with all the space-view terminology breaking changes (`SpaceView`->`View`, `space_view`->`view`, etc).

crates/store/re_chunk_store/src/writes.rs

@@ -51,6 +51,16 @@ impl ChunkStore {
 
         let mut chunk = Arc::clone(chunk);
 
+        // TODO(#6889): This is bad for performance, but indicators should go away all together soon anyways?
+        if chunk
+            .components()
+            .keys()
+            .any(|descr| descr.component_name.is_indicator_component())
+        {
+            let patched = chunk.patched_weak_indicator_descriptor_023_compat();
+            chunk = Arc::new(patched);
+        }
+
         if self.id.kind == re_log_types::StoreKind::Blueprint {
             let patched = chunk.patched_for_blueprint_021_compat();
             chunk = Arc::new(patched);