v0.3-T4 — ahc simulate (dry-run with cost projections)

[sv-pro]

Summary

Implement ahc simulate <manifest> <trace-or-scenario> — dry-run a trace or scenario set against a manifest without executing real tools. Output: decision table (ALLOW/DENY/ASK per action) with p50/p90 cost projection per step.

Depends on: #121 (validate), #123 (cost-estimate)

Scope

• Parse trace files or scenario YAML; replay each action through compiled policy without touching real tools
• Output: per-action decision table with ALLOW/DENY/ASK verdict + cost estimate
• Simulation fidelity invariant: same decisions as live runtime for the reference scenario set
• --compare <live-trace> flag to diff simulation decisions against recorded live decisions

Key Files

• src/agent_hypervisor/compiler/cli.py — add simulate command
• src/agent_hypervisor/runtime/executor.py — SimulationExecutor (exists, extend)
• tests/compiler/test_simulate.py — new test file; include fidelity test against workspace manifest

Acceptance Criteria

• ahc simulate manifests/workspace_v2.yaml <scenario> outputs decision table
• Output includes cost projection (p50/p90) per step
• Simulation fidelity: decisions match live runtime for reference workspace scenario set
• pytest tests/compiler/test_simulate.py -v passes

Milestone

v0.3