TMSS-1124: create a subtask_output per task_connector_type of the task, but only with role Correlator or Beamformer because that is what observations produce

TMSS-1124: use newly implemented SubtaskInput/Output to filter and connect the correct dataproducts between subtasks

TMSS-1117: use transactions. If one task/subtask creation/update fails, then rollback the whole transaction

TMSS-920: reverted erroneously changed cluster_name for observations. In this ticket we only use the cluster_resources for pipelines.

TMSS-920: Specify parallellism in nr of tasks for backwards compatibility, and group cluster & partition explicitly together.

TMSS-920: Propagate cluster_resource settings task->subtask and subtask->parset. Added cluster_resources key to pipeline subtask specifications.

TMSS-159: further adaptations to new start/stop times, add aggregate time properties, set process times when scheduling

TMSS-850: use json add_defaults_to_json_object_for_schema/get_default_json_object_for_schema from common Template class

TMSS-917: ensure that each parent task_blueprint has exactly one primary child subtask; added and adapted tests

TMSS-917: now that we reverted to a single parent TaskBlueprint for a Subtask there is no need anymore for a SubtaskOutput to have a task_blueprint reference.

TMSS-917: refactored the parallel LBA calibrator and target observations into one task with one specification with two parts (for the calibrator and the target). This enables us to have a single parent task for each subtask again, which is preferrable

TMSS-872: Moved subtasks.py back to main dir, to prevent merging issues with refactoring happening in other branches.