Cluster-wide Auto Tune backoff/notification config

Cluster-wide backoff policy — the "when a change misbehaves" guardrail applying to every enabled optimization on the cluster.

Stored under the backoff_policy key in databricks_auto_tune_configuration.configuration. The integer defaults come straight from the backoff monitor's own env-overridable constants (backoff_constants.py) so the form's baselines always match what the pod actually enforces. Execution time and query latency are deliberately absent — neither signal applies to all-purpose clusters (no per-run baseline, no query stream). Each threshold belongs to one optimization axis and is editable only while that axis is enabled: restarts_after_shutdown to Auto Termination; cold_start_threshold_per_day to Min Worker Node Sizing. The per-step monitoring horizon is no longer configurable — each axis holds a fixed backoff window (_window_days_for_axis in the monitor), so the step-down cadence is not user-tunable.

Cluster-wide Auto Tune notification preferences.

Stored under the notification_config key in databricks_auto_tune_configuration.configuration. References Slack channel ids from the tenant's notifications table by reference only — never a webhook URL inline (see NotificationConfigPublic for the same redaction rule on the generic notifications settings page).