Refactor: Auto-inject Agent State Schemas and simplify agent configuration API

[Cyb3rWard0g]

Overview:

This PR eliminates unsafe user-facing schema customization by auto-injecting the correct state model bundle based on agent/orchestrator type. Users now provide only storage configuration, and the framework automatically selects the appropriate schemas (AgentWorkflowState for agents, LLMWorkflowState for orchestrators), preventing runtime AttributeError crashes from incompatible custom schemas.

Key Changes:

• Introduced StateModelBundle abstraction to encapsulate state schemas and hook functions with validation guards
• Auto-injection pattern where agent/orchestrator classes inject their required schema bundle during initialization without user intervention
• Simplified AgentStateConfig API by removing exposed state_model_cls and message_model_cls fields that allowed unsafe customization
• Three-tier fallback resolution for state hooks: config overrides → default bundle → hardcoded defaults
• Added type hints for default_bundle parameter using TYPE_CHECKING to prevent circular imports
• Fixed incorrect import paths in LLM orchestrator compensation logic (dapr_agents.workflow.orchestrators → dapr_agents.agents.orchestrators)
• Updated all quickstarts to use simplified configuration syntax (removed pubsub_config=, state_config= prefixes)
• Added test coverage for bundle injection and conflict detection

[sicoyle]

thank you for your efforts on this! I love the idea of having the state implementation unified for agents and orchestrators and make things easier on us to maintain 🙌 🌟

few comments so far, but yeah I'm curious for your thoughts on. What are the benefits of the factory/coercer setup vs just having a static state class for agents and another for orchestrators that are slightly diff but use the same underlying storage base class? Would that be easier than this approach?

[sicoyle]

pls add the inline comment on the github comment above and then i am good with this one. Thank youuuu!!