fix: use warning logging for retries

[MatufA]

Summary

• replace direct stderr retry prints with secops.auth warning logs
• preserve the existing f-string retry messages and caller-controlled logging
• cover response-triggered and error-triggered retries without stderr output

Why

Raw stderr retry messages can be buffered with neighboring application logs
before Datadog ingestion. Because those messages bypass Python logging, they
lack consistent level and logger metadata and can cause downstream parser
discrepancies.

Validation

• pytest tests/ -m "not integration" -k "not test_time_config": 772 passed,
  208 deselected, 64% coverage
• pylint --rcfile=pylintrc src/secops/auth.py: passed
• black --check --line-length=80 --target-version=py310 src/secops/auth.py:
  passed
• manual LogRetry driver: both retry paths produced separate warning records

The unrelated test_time_config unit test was excluded locally because it
attempts to create ~/.secops, which is blocked by the execution sandbox.
Integration tests remain code-owner-triggered per TESTING.md.

Fixes #221