feat: infracost v2 support — deployml estimate and deployml costs

[andrewtclim]

Summary

• Rewrites infracost.py for infracost v2 (infracost scan / infracost inspect --json — the old infracost breakdown command no longer exists in v2 and was silently failing)
• Adds deployml estimate: pre-deploy cost prediction from config.yaml, no GCP credentials or infra needed
• Adds deployml costs: cost check against an existing deployed Terraform workspace
• Adds deployml doctor auth check (infracost v2 requires an account + infracost auth login)
• Fixes copy_modules_to_workspace to always include cloud_sql_postgres module when mlflow uses postgresql backend (was silently returning $0)
• 14 unit tests for infracost.py
• Simplified docs in costs.md and the Cloud Run tutorial

What the student experience looks like

# Before deploying — no GCP credentials needed
deployml estimate
# → Monthly Cost: $34.55 USD | Cloud SQL: $34.55

# Inline during deploy, before confirmation
# → Deploy stack? Monthly cost: ~$34.55 USD [y/N]:

# After deploying — reads real deployed workspace
deployml costs
# → Monthly Cost: $34.55 USD | Cloud SQL: $34.55

Limitations / future work

deployml estimate and deployml costs read Terraform configs via infracost — accurate for always-on resources (Cloud SQL, VMs) but show $0 for usage-based services (BigQuery, GCS, Cloud Run). A future deployml billing command could query the GCP Cloud Billing API for actual accrued charges.

Test plan

• pytest tests/test_infracost.py — 14 tests, all passing
• deployml doctor — shows Infracost installed + authenticated
• deployml estimate — shows cost from config.yaml without deploying
• deployml deploy — cost shown inline before confirmation prompt
• deployml costs — shows cost of running deployment
• deployml costs with no deployment — helpful error message

🤖 Generated with Claude Code