epic: adaptive execution architecture — four models, context bridging, lineage (ADR-0007)

[mdproctor]

Summary

Implements the full architecture defined in ADR-0007: four co-equal execution models feeding a shared execution kernel, pluggable context bridging from CaseContext through WorkflowContext to AgenticScope and beyond, complete causal lineage across all nested boundaries, and frictionless langchain4j-agentic integration.

Design References

• ADR-0007: `adr/0007-four-execution-models-as-first-class-citizens.md`
• Design discussion: design: true hybrid execution — choreography, orchestration (rules + workflow), and plans #200
• langchain4j agents: https://github.com/langchain4j/langchain4j/blob/main/docs/docs/tutorials/agents.md
• Quarkus Flow ↔ AgenticScope: https://github.com/quarkiverse/quarkus-flow/blob/main/docs/modules/ROOT/pages/concepts-agentic-langchain4j.adoc

Child Epics

Epic	Area
#202	Agnostic routing layer and execution kernel
#203	ContextBridge protocol and worker-level context selection
#204	Scope propagation, return-path tracking, and write-through
#205	Complete causal lineage across all execution boundaries
#206	Orchestration — workflow-based (FlowWorker ↔ WorkOrchestrator)
#207	Orchestration — rules-based (Drools/DMN → WorkOrchestrator)
#208	Plan-based execution (PlanExecutor, Plan model, PlanSource SPI)
#209	langchain4j-agentic integration (AgenticScopeBridge, CasehubPlanner, AgentListener)
#210	Cancellation, timeout, and error recovery across boundaries
#211	Observability — lineage queries, context stack visibility, A2A tracing

Guiding Principles (from ADR-0007)

1. All four execution models share one execution kernel. Only the router differs.
2. No execution model is privileged — choreography is not the default.
3. Context model is worker-level, not case-level. Null bridge = CaseContext directly (backward compatible).
4. Every unit of work at every level emits a `CaseLedgerEntry`. `causedByEntryId` chains are complete.
5. langchain4j: `AgenticScope` ↔ `WorkflowContext` ↔ `CaseContext` — three distinct contexts, bridged not merged.
6. `ContextBridge` SPI makes future execution models and context types pluggable.

Suggested implementation order

1. epic: agnostic routing layer and execution kernel #202 Routing kernel (foundation for everything)
2. epic: ContextBridge protocol and worker-level context selection #203 ContextBridge (foundation for context work)
3. epic: scope propagation, return-path tracking, and context write-through #204 Scope propagation (enables correct multi-level context flow)
4. epic: complete causal lineage across all execution boundaries #205 Lineage (enables observability on all work done above)
5. epic: orchestration — workflow-based (FlowWorker ↔ WorkOrchestrator bridge) #206 Workflow orchestration (largest gap, highest demand)
6. epic: plan-based execution — PlanExecutor, Plan data model, PlanSource SPI #208 Planning (enables LLM-driven execution)
7. epic: langchain4j-agentic integration — AgenticScopeBridge, CasehubPlanner, AgentListener #209 langchain4j integration (builds on all of the above)
8. epic: orchestration — rules-based (Drools/DMN → WorkOrchestrator) #207 Rules orchestration (lower priority, existing primitives mostly cover it)
9. epic: cancellation, timeout, and error recovery across execution boundaries #210 Cancellation/error recovery (needed for production hardening)
10. epic: observability — cross-boundary lineage queries, context stack visibility, A2A tracing #211 Observability (builds on epic: complete causal lineage across all execution boundaries #205; progressive enhancement)