v0.49.0

• Setting both projectVpcId and projectVPCRef is now rejected when the resource is applied, also
  when webhooks are disabled. A resource that already sets both must drop one before its next update
• Change Go type of BaseServiceFields.ProjectVPCID from string to *string and the JSON tag of
  Tags and TechnicalEmails from omitempty to omitzero. CRD schema and manifests are
  unchanged; only Go code importing api/v1alpha1 is affected
• Fix service webhooks refusing to delete a resource that sets both projectVpcId and projectVPCRef;
  the conflict is still rejected on create and update
• Add Kafka field userConfig.kafka.group_consumer_heartbeat_interval_ms, type integer: The heartbeat
  interval given to the members of a consumer group using the consumer rebalance protocol
• Add Kafka field userConfig.kafka.group_consumer_max_heartbeat_interval_ms, type integer: The
  maximum heartbeat interval allowed for consumer group members using the consumer rebalance protocol
• Add Kafka field userConfig.kafka.group_consumer_max_session_timeout_ms, type integer: The maximum
  session timeout allowed for consumer group members using the consumer rebalance protocol
• Add Kafka field userConfig.kafka.group_consumer_min_heartbeat_interval_ms, type integer: The
  minimum heartbeat interval allowed for consumer group members using the consumer rebalance protocol
• Add Kafka field userConfig.kafka.group_consumer_min_session_timeout_ms, type integer: The minimum
  session timeout allowed for consumer group members using the consumer rebalance protocol
• Add Kafka field userConfig.kafka.group_consumer_session_timeout_ms, type integer: The timeout
  used to detect consumer group member failures when using the consumer rebalance protocol
• Add MySQL field userConfig.mysql.innodb_max_dirty_pages_pct_lwm, type number: Percentage of dirty
  pages in the InnoDB buffer pool at which InnoDB starts pre-flushing, to keep the dirty page ratio
  from reaching innodb_max_dirty_pages_pct
• Add MySQL field userConfig.mysql.innodb_max_dirty_pages_pct, type number: Percentage of dirty
  pages in the InnoDB buffer pool at which InnoDB starts flushing aggressively
• Add MySQL field userConfig.mysql.innodb_max_purge_lag_delay, type integer: Cap in microseconds
  on the per-row delay innodb_max_purge_lag applies to INSERT, UPDATE and DELETE
• Add MySQL field userConfig.mysql.innodb_max_purge_lag, type integer: Number of outstanding transactions
  the InnoDB purge operation may fall behind by before INSERT, UPDATE and DELETE are delayed
  to let it catch up
• Add OpenSearch field userConfig.opensearch.ml_commons_max_model_on_node, type integer: Maximum
  number of ML models that can be deployed on a single ML node. Defaults to 10
• Add OpenSearch field userConfig.opensearch.ml_commons_model_auto_deploy_enable, type boolean:
  For externally hosted models only: automatically deploy a model on its first predict request instead
  of requiring an explicit deploy call first
• Add OpenSearch field userConfig.opensearch.ml_commons_model_auto_redeploy_enable, type boolean:
  Automatically redeploy deployed or partially deployed models after ML nodes rejoin the cluster,
  e.g
• Change PostgreSQL field userConfig.variant: enum add pg_lake
• Add ServiceIntegration field datadog.datadog_pg_dbname, type string: Database the Datadog PostgreSQL
  check connects to
• Add Valkey field userConfig.replicas, type integer: Number of replicas per shard in the Valkey
  cluster. Only applies to cluster plans
• Add Valkey field userConfig.shard_count, type integer: Number of shards (primaries) in the Valkey
  cluster. Required for cluster plans
• Add KafkaConnect field userConfig.custom_plugins, type array: Install custom plugins uploaded
  via the custom file service
• Add connInfoSecretRoute (dynamic, public, private, privatelink) to ServiceUser and
  ServiceUserRotation. It selects the service endpoint published to the connection Secret as
  HOST and PORT; for Kafka also SASL_HOST/SASL_PORT and SCHEMA_REGISTRY_HOST/SCHEMA_REGISTRY_PORT.
  Set privatelink to publish the PrivateLink endpoint. Until the service exposes the chosen route,
  the Secret is left untouched and the resource reports an Error condition with reason
  Preconditions; a ServiceUserRotation also pauses rotation and runs as soon as the route appears.
  Leaving the field unset keeps the behavior of previous releases, so existing Secrets are not rewritten on upgrade.
• Fix Reconciler error ("the object has been modified") logged when a resource changed while the
  operator was writing its status. The conflict is now ignored and the state is recomputed by the next reconcile.
• Add OpenSearch field userConfig.opensearch.knn_cache_item_expiry_enabled, type boolean: Enable
  or disable removing KNN graphs from the cache when they are idle. Disabled by default
• Add OpenSearch field userConfig.opensearch.knn_cache_item_expiry_minutes, type integer: If enabled,
  the amount of time in minutes a KNN graph can be idle in the cache before being evicted
• Add OpenSearch field userConfig.opensearch.knn_circuit_breaker_unset_percentage, type number:
  The native memory usage, as a percentage below the KNN memory circuit breaker limit, at which a tripped
  circuit breaker resets
• Fix resources staying ReadyToUse after their connection secret failed to publish. The resource is
  now reported as not ready until the secret is published successfully.