revert: remove Azure ML parallelization (PR #24)

[abrichr]

Summary

Reverts #24 - Azure ML parallelization for WAA evaluation.

Reason

We're using the direct VM pool approach instead (pool-create, pool-wait, pool-run in openadapt-ml). The Azure ML approach adds complexity without clear benefits:

• Azure ML requires ACR setup and authentication
• Job management is more complex than direct SSH
• VM pools are simpler and working

The reverted code is ~2800 lines of dead code that would cause confusion.

What was removed

• Azure ML job submission/management
• Azure ML compute instance management
• Azure ML log fetching
• Related CLI commands (azure-ml-*)

🤖 Generated with Claude Code