feat: migrate evaluation infrastructure from openadapt-ml (#29)

* feat: migrate evaluation infrastructure from openadapt-ml

Move all evaluation infrastructure (~13,000 lines) from openadapt-ml/benchmarks/
to openadapt-evals so openadapt-ml can focus on pure ML (schemas, training,
inference, model adapters).

Migrated modules:
- benchmarks/vm_cli.py: Full VM/pool CLI with 50+ commands (8,503 lines)
- infrastructure/azure_vm.py: AzureVMManager with SDK + CLI fallback
- infrastructure/pool.py: PoolManager for multi-VM orchestration
- infrastructure/resource_tracker.py: Azure cost tracking
- benchmarks/pool_viewer.py: Pool results HTML viewer
- benchmarks/trace_export.py: Training data export (keeps openadapt_ml.schema dep)
- waa_deploy/: Docker agent deployment files

Also adds:
- config.py: Pydantic-settings config for Azure credentials
- pydantic-settings + azure-mgmt-* dependencies
- 4 test files migrated from openadapt-ml

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

* fix: correct DOCKERFILE_PATH and stale debug path in vm_cli

- DOCKERFILE_PATH: use parent.parent to reach waa_deploy/ from benchmarks/
- cmd_tail_output: update hardcoded task dir from openadapt-ml to openadapt-evals

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

---------

Co-authored-by: Claude Opus 4.6 <noreply@anthropic.com>

Author: abrichr

openadapt_evals/benchmarks/pool_viewer.py

@@ -0,0 +1,51 @@
+from __future__ import annotations
+
+from pydantic_settings import BaseSettings
+
+
+class Settings(BaseSettings):
+    """Application settings loaded from environment variables or .env file.
+
+    Priority order for configuration values:
+    1. Environment variables
+    2. .env file
+    3. Default values (None for API keys)
+    """
+
+    # VLM API Keys
+    anthropic_api_key: str | None = None
+    openai_api_key: str | None = None
+    google_api_key: str | None = None
+
+    # Azure credentials (for WAA benchmark on Azure)
+    # These are used by DefaultAzureCredential for Service Principal auth
+    azure_client_id: str | None = None
+    azure_client_secret: str | None = None
+    azure_tenant_id: str | None = None
+
+    # Azure ML workspace config
+    azure_subscription_id: str | None = None
+    azure_ml_resource_group: str | None = None
+    azure_ml_workspace_name: str | None = None
+
+    # Azure resource group for VM operations (used by benchmarks CLI)
+    azure_resource_group: str = "openadapt-agents"
+
+    # Azure VM settings (optional overrides)
+    azure_vm_size: str = "Standard_D2_v3"
+    azure_docker_image: str = "docker.io/windowsarena/winarena:latest"
+
+    # Azure Storage for async inference queue
+    azure_storage_connection_string: str | None = None
+    azure_inference_queue_name: str = "inference-jobs"
+    azure_checkpoints_container: str = "checkpoints"
+    azure_comparisons_container: str = "comparisons"
+
+    model_config = {
+        "env_file": ".env",
+        "env_file_encoding": "utf-8",
+        "extra": "ignore",  # ignore extra env vars
+    }
+
+
+settings = Settings()

openadapt_evals/benchmarks/trace_export.py

@@ -1,32 +1,39 @@
 """Infrastructure components for VM management and monitoring.
 
 This module provides:
+- AzureVMManager: Azure VM lifecycle management (SDK + CLI fallback)
+- PoolManager: Multi-VM pool orchestration
 - VMMonitor: Azure VM status monitoring
 - AzureOpsTracker: Azure operation logging
 - SSHTunnelManager: SSH tunnel management for VNC/API access
 
 Example:
     ```python
-    from openadapt_evals.infrastructure import VMMonitor, SSHTunnelManager
+    from openadapt_evals.infrastructure import AzureVMManager, PoolManager
 
-    # Monitor VM status
-    monitor = VMMonitor()
-    status = monitor.get_status()
+    # Manage VMs
+    vm = AzureVMManager()
+    ip = vm.get_vm_ip("waa-eval-vm")
 
-    # Manage SSH tunnels
-    tunnel_manager = SSHTunnelManager()
-    tunnel_manager.start_tunnels_for_vm("172.171.112.41", "azureuser")
+    # Create and manage pools
+    pool = PoolManager()
+    pool.create(workers=3)
     ```
 """
 
-from openadapt_evals.infrastructure.vm_monitor import VMMonitor, VMConfig
 from openadapt_evals.infrastructure.azure_ops_tracker import AzureOpsTracker
+from openadapt_evals.infrastructure.azure_vm import AzureVMManager
+from openadapt_evals.infrastructure.pool import PoolManager, PoolRunResult
 from openadapt_evals.infrastructure.ssh_tunnel import SSHTunnelManager, get_tunnel_manager
+from openadapt_evals.infrastructure.vm_monitor import VMMonitor, VMConfig
 
 __all__ = [
+    "AzureOpsTracker",
+    "AzureVMManager",
+    "PoolManager",
+    "PoolRunResult",
     "VMMonitor",
     "VMConfig",
-    "AzureOpsTracker",
     "SSHTunnelManager",
     "get_tunnel_manager",
 ]