fix(azure): retry transient VM inventory disappearance

[ruby-dlee]

Summary

• retry only explicit ResourceNotFound from the read-only az vm list --show-details expansion race
• keep all other inventory errors, malformed responses, identity checks, and mutation paths fail-closed
• bound retry attempts to four with short backoff

Live evidence

Concurrent official cleanup deleted a VM while Azure CLI expanded instance views. The CLI returned ResourceNotFound for the whole inventory even though this was a truthful transient read race. Slots 9 and 10 subsequently reached compute-removed.

Verification

• bash tests/fm-worker-lifecycle.test.sh
• python3 -m py_compile bin/fm-azure-worker-provider.py
• ShellCheck on changed shell test
• git diff --check
• cold adversarial review: CLEAR

Reviewed source: 40e4a84a9f6d173644935a512380076dff37da4e