Allow disabling or removing the default namespace index of an event source's informer

Beginne damit, ManagedInformerEventSource über InformerManager und InformerWrapper hinweg nachzuverfolgen, und untersuche anschließend InformerConfiguration sowie die vorhandenen Einstiegspunkte AbstractInformerPool und ConfigurationServiceOverrider. Vergleiche diese Pfade mit SharedIndexInformer.removeNamespaceIndex(). Die Aufgabe ist abgeschlossen, wenn eine erstklassige Konfigurationsoption den standardmäßigen Namespace-Index deaktivieren kann, ohne experimentelles Subclassing des Pools zu erfordern, und das Startverhalten des Informers abgedeckt ist.

For InformerEventSources, fabric8's default "namespace" index is unused: InformerWrapper.list(namespace) filters the item store directly, and byIndex only serves the JOSDK-qualified custom indexes. The index still costs one entry per object in the watched scope, and when an event source uses a filtering ItemStore to bound its cache, this unused index becomes the dominant retained memory (compounded by the fabric8 cleanup gap reported in fabric8io/kubernetes-client#8103).

fabric8 already exposes SharedIndexInformer.removeIndexer(String), but the SDK offers no path to the informer (it sits behind ManagedInformerEventSource's InformerManager and InformerWrapper, all inaccessible), so today the only way is reflection over three private fields (see the edit below, the experimental informer pool provides a path).

Suggestion: an InformerConfiguration flag (e.g. withoutNamespaceIndex()) applied when the wrapper starts the informer, or an extension point that hands advanced users the SharedIndexInformer for tuning.

Would you be open to a PR for this, and which of the two shapes would you prefer?

EDIT: found out that overriding AbstractInformerPool.start (via ConfigurationServiceOverrider.withInformerPool) allows calling SharedIndexInformer.removeNamespaceIndex() before the informer starts, so reflection is not required after all. The API is marked experimental though, and subclassing the pool for a per-informer concern is indirect, so we think a first-class InformerConfiguration option is still worth it. Happy to include that context in the PR.