feat: proactive stack health reconciliation and periodic Docker resource pruning

[veerendra2]

Closes #31

Summary

• Health reconciliation loop (HEALTH_RECONCILE_INTERVAL, default 5m): periodically checks all managed stacks and redeploys any where Status != Running or any container is exited/dead. Skips stacks with restarting containers (Docker restart policy is doing its job).
• Periodic Docker resource prune (PRUNE_INTERVAL, default 24h): prunes unused images, volumes, and build cache — but only when all managed stacks are healthy. If any stack is degraded or suspended, the cycle is skipped.
• Suspend label (composeflux.suspend=true): set via docker update --label-add on any container to pause reconciliation for the entire stack. Both the health loop and prune loop respect it. Useful during maintenance (e.g., stopping a DB for backups without triggering a reconcile).
• Moved inline dClient.Prune() out of Prune() into the dedicated scheduled loop so pruning is always gated on health.

How to Test

Health reconciliation:

# Start composeflux with a short interval for faster testing
HEALTH_RECONCILE_INTERVAL=30s docker compose up -d

# Stop a container inside a managed stack
docker stop <container>

# Within 30s, verify it comes back up
docker ps | grep <container>

Suspend label:

# Set suspend label on a container
docker update --label-add composeflux.suspend=true <container>

# Stop another container in the same stack
docker stop <another-container-in-same-stack>

# Confirm the health loop does NOT restart it (check logs)
docker logs composeflux | grep -i suspend

Prune loop:

# Start with a short prune interval
PRUNE_INTERVAL=2m PRUNE_RESOURCES=true docker compose up -d

# With all stacks healthy — after 2m check logs for prune firing
docker logs composeflux | grep -i prune

# Stop a managed stack, wait for next tick — confirm prune is skipped
docker compose -p <stack> stop
docker logs composeflux | grep "Skipping Docker resource prune"

🤖 Generated with Claude Code