Split delivery and run state interfaces

[darthfork]

Summary

• split production state consumption into delivery, workflow-run, and workflow-job interfaces
• wire webhook ingestion to delivery state only and run transitions to their specific run-state backends
• keep Redis as the shared concrete persistence adapter while avoiding a combined production StateStore interface

TDD notes

• RED: added interface-level coverage showing default webhook ingestion can use delivery state without run state
• RED: added workflow run/job metric tests proving run transitions can use their own state interfaces without delivery state or the other run state
• GREEN/REFACTOR: introduced narrow state backends, adapted run transitions, and updated integration helpers to pass state dependencies independently

Cardinality guardrails

• no Prometheus collector definitions in src/metrics.go were changed
• metric recorder cardinality guard tests remain in place

Verification

• make test-all
• make lint
• make security

Author

• Codex (gpt-5.5)

[darthfork]

Good design — ISP applied correctly here. The three narrow backends replace a fat interface cleanly, and the failingDeliveryStore simplification is a nice side-benefit (no more embedded inMemoryStateStore that can bleed state). A few things worth tightening before merge.