Phase 81: expose CodingPod refactorer API

[pcharbon70]

Summary

• expose AgentWorkspace.refactor_work/3,4 through the existing per-work-item CodingPod lifecycle, lazy :refactorer node, task-board artifacts, pod metadata, LLM selection, semantic/memory context, and workflow provenance capture
• add Phase 81 planning and developer guidance for product-owned refactorer routing while keeping full_workflow/3,4 as plan -> code -> review
• add focused workspace and CodingPod integration coverage for refactorer metadata, isolated work-item pods, teardown, and workflow-provenance patch capture
• carry forward the unmerged conversation fixes from this work branch: EventRecord append accept cleanup, after-commit conversation persistence notifications, and duplicate active work-item conversation rejection

Verification

• mix test test/jido_code/agent_workspace_test.exs test/jido_code/agent_os_integration_test.exs --max-failures 1
• mix memory.verify
• mix source_graph.verify
• mix compile --warnings-as-errors
• mix test test/jido_code/conversations_coordinator_test.exs --max-cases 1 --max-failures 1
• mix test test/jido_code/conversations_driver_test.exs --max-cases 1 --max-failures 1
• mix test test/jido_code/conversations_test.exs --max-cases 1 --max-failures 1
• mix test test/jido_code/conversations_pubsub_test.exs --max-cases 1 --max-failures 1
• mix test test/jido_code/conversations/context_memory_test.exs --max-cases 1 --max-failures 1

Known existing issue: the combined conversation-runtime batch still fails when those conversation files are run together, reproducing the existing Conversations.ChildSupervisor shutdown/no-process failure in test/jido_code/conversations_coordinator_test.exs:448. The same coordinator test and the involved conversation files pass individually.

Note: bw prime and bw ready were attempted before work, but bw is not installed in this environment.