Simulation v1 to v2

Simulation v2 applies the Blueprint pattern to tests. In v1, test cases were authored and owned in Hamming, hand-written per team, and Avoca held little more than a pointer to the results. In v2, the test definitions live in Avoca's database as templates carrying generation instructions and focus areas; at run time the generate job compiles a case, merges in the team's live prompt variables (the same variable layer the agent itself compiles from), and pushes it up to Hamming, which generates the concrete scenarios and executes them as real voice calls. Avoca becomes the source of truth and Hamming becomes the real-time executor. Our contribution to v2 is the transfer mock: on a test call the handler resolves the real destination, records it, narrates it, and skips the dial, which makes transfer routing (previously refused outright on test calls, and therefore structurally untestable) both safe and observable.