Improve fragile input state fetching #25358

Description

We currently rely on GET /cluster/inputstates, which performs HTTP fan-out across all nodes and returns full InputStateSummary objects. This is expensive, fragile in larger clusters, and prone to timeouts or node overload.

This issue tracks the migration to DB-backed input state handling, now that input states are stored in MongoDB (input_runtime_states).

Goal

Incrementally remove the fan-out-based system and migrate all consumers to DB-backed endpoints:

• Add a lightweight input state summary endpoint
• Move InputRuntimeStatusProvider to the DB-based approach
• Remove workaround logic currently present in InputRegistry
• Add detailed per-node state endpoint
• Decouple input definitions from state fetching
• Fully remove GET /cluster/inputstates and related code paths