deprovision: firewall deletion fails with resource_in_use immediately after server delete

Description

During node deprovisioning, the Hetzner firewall DELETE call fails with a 422 resource_in_use error immediately after the server is confirmed deleted. Re-running deprovision a few seconds later succeeds (server skipped as already gone, firewall deletes cleanly).

Steps to Reproduce

1. Run deprovision on a provisioned node
2. Server deletion completes and is confirmed
3. Firewall deletion immediately follows — returns 422 resource_in_use

Expected Behavior

Firewall deletion should succeed in a single deprovision run.

Actual Behavior

[NP-E021] Hetzner firewall deletion failed: DELETE /firewalls/{id} returned 422: resource_in_use

The error is transient — re-running deprovision after a few seconds always succeeds.

Root Cause (suspected)

Hetzner's server deletion is async. Even after the server deletion is confirmed, the firewall detachment takes a moment to propagate on Hetzner's side. The deprovision method proceeds to firewall deletion too quickly.

Suggested Fix

Add a short retry loop with backoff (e.g. 3 attempts, 3s apart) around the firewall DELETE call, or add an explicit wait after server deletion before attempting firewall cleanup.