InstanceHA production readiness improvements

- Remove proactive Nova re-auth on every poll cycle; rely on existing
  error-path reconnection (saves 2-3 HTTPS round-trips per 45s cycle)
- Fix _is_service_stale timezone parsing for Python 3.9 (handle Z suffix,
  remove dead isinstance(datetime) branch)
- Sanitize exception messages in K8s Events to prevent credential leaks
  via namespace-readable Event objects
- Add Authorization header pattern to _SECRET_PATTERNS and consistently
  sanitize exception strings in _retry_with_backoff, _monitor_evacuation,
  and _server_evacuate_future
- Reuse requests.Session() for Redfish power-off polling loop to avoid
  ~30 redundant TLS handshakes per fencing operation
- Add 5-second client timeout to HealthHandler to prevent slow-client
  denial of liveness probes
- Remove dead _host_servers_cache attribute
- Set service.ready=False and signal UDP listener stop events in SIGTERM
  handler for clean shutdown

Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>

Author: lmiccini