fix(azure): tolerate child disappearance during inventory

[ruby-dlee]

Summary

• treat explicit Azure ResourceNotFound as truthful absence only for child-detail reads performed during inventory
• keep every mutation, cleanup, execute, and retirement caller fail-closed by default
• preserve malformed-success and all other provider-error refusals

Live failure addressed

Concurrent exact cleanup removed a TTL schedule after another inventory had listed it but before that inventory fetched the full child. The other claims failed closed even though no foreign resource or identity mismatch existed.

Verification

• python3 -m py_compile bin/fm-azure-worker-provider.py
• shellcheck --norc -x tests/fm-worker-lifecycle.test.sh
• git diff --check
• bash tests/fm-worker-lifecycle.test.sh
• cold adversarial review: CLEAR

Deployment will pin the exact reviewed source revision; no updater is involved.