Stabilize SaaS lifecycle polling and API error handling

[tkilias]

Summary

Stabilize the handwritten SaaS lifecycle helpers around database deletion, allowed-IP polling, and malformed backend error payloads.

Problem

The current helper behavior is brittle in several transitional backend states:

• database deletion polling can observe stale or intermediate states
• allowed-IP lifecycle polling can rely on stale list results
• timeout values are too short for some real integration runs
• malformed backend error payloads can break error handling when ApiError fields are missing

These problems make the related tests flaky and complicate diagnosis when failures happen in CI.

Proposed change

• harden database deletion polling
• harden allowed-IP polling
• increase the relevant timeout defaults
• keep debug logging for API helper responses useful during diagnosis
• tolerate malformed backend ApiError payloads in the generated model

Impact

This should reduce flaky lifecycle checks and make failures easier to diagnose without changing the public API shape.

Related PR

• [codex] Fix failing tests and timeouts #168