feat: remove evaluation infrastructure (moved to openadapt-evals)

[abrichr]

Summary

• Removes ~18,000 lines of evaluation infrastructure that has been migrated to openadapt-evals (PR feat: migrate evaluation infrastructure from openadapt-ml openadapt-evals#29)
• openadapt-ml is now a pure ML package: schemas, training, inference, model adapters, and ML-specific benchmark agents
• Removes azure-ai-ml, azureml-core, and azure-mgmt-* from dependencies

Deleted files

• benchmarks/cli.py (8,503 lines) - VM/pool management CLI
• benchmarks/azure_vm.py - AzureVMManager
• benchmarks/pool.py - PoolManager
• benchmarks/vm_monitor.py, azure_ops_tracker.py, resource_tracker.py
• benchmarks/azure.py, viewer.py, pool_viewer.py, trace_export.py
• benchmarks/waa_deploy/ - Docker agent deployment
• 4 test files moved to openadapt-evals

Kept in openadapt-ml

• benchmarks/agent.py - PolicyAgent, APIBenchmarkAgent, UnifiedBaselineAgent (ML model wrappers)

Migration Guide

Old import (openadapt-ml)	New import (openadapt-evals)
openadapt_ml.benchmarks.cli	openadapt_evals.benchmarks.vm_cli (or oa-vm CLI)
openadapt_ml.benchmarks.azure_vm.AzureVMManager	openadapt_evals.infrastructure.azure_vm.AzureVMManager
openadapt_ml.benchmarks.pool.PoolManager	openadapt_evals.infrastructure.pool.PoolManager
openadapt_ml.benchmarks.vm_monitor.VMMonitor	openadapt_evals.infrastructure.vm_monitor.VMMonitor
openadapt_ml.benchmarks.azure_ops_tracker	openadapt_evals.infrastructure.azure_ops_tracker
openadapt_ml.benchmarks.resource_tracker	openadapt_evals.infrastructure.resource_tracker
openadapt_ml.benchmarks.pool_viewer	openadapt_evals.benchmarks.pool_viewer
openadapt_ml.benchmarks.trace_export	openadapt_evals.benchmarks.trace_export
openadapt_ml.benchmarks.waa_deploy	openadapt_evals.waa_deploy

CLI migration

Old command	New command
python -m openadapt_ml.benchmarks.cli pool-create	oa-vm pool-create
python -m openadapt_ml.benchmarks.cli pool-run	oa-vm pool-run
python -m openadapt_ml.benchmarks.cli pool-status	oa-vm pool-status
python -m openadapt_ml.benchmarks.cli pool-cleanup	oa-vm pool-cleanup
python -m openadapt_ml.benchmarks.cli create	oa-vm create
python -m openadapt_ml.benchmarks.cli status	oa-vm status
python -m openadapt_ml.benchmarks.cli vm monitor	oa-vm vm monitor
All other CLI commands	oa-vm <command>

What stays the same

# ML agents - unchanged
from openadapt_ml.benchmarks import PolicyAgent, APIBenchmarkAgent, UnifiedBaselineAgent

Test plan

• uv run pytest tests/ -v — 253 passed, 6 skipped
• from openadapt_ml.benchmarks import PolicyAgent, APIBenchmarkAgent, UnifiedBaselineAgent works
• ruff check passes
• Verify no remaining references to deleted modules in non-deleted code

🤖 Generated with Claude Code