Ayeaye

Changed
- Model's use of external logger to instead support multiple external loggers in the same as Fossa does

Fixed
- hung if PartitionedModel.partition_slice returns an empty list
- stray bracket resulting in a couple of unittests being skipped

Added
- optional task_id and failure_origin_task_id fields to ayeaye.runtime.task_message.TaskFailed. Theseare useful to Fossa

Added
- A PartitionedModel can run other PartitionedModel(s)

Changed
- LocalProcessPool.run_subtasks no longer creates daemon Processes. This makes it possible to a Par
tionedTask to spawn other Partitioned tasks. The downside is the potential for orphaned subtasks.

Fixed
- AbstractProcessPool.run_subtasks's context_kwargs should be a plain dictionary and doesn't need the 'mapper' key from :meth:`ayeaye.ayeaye.connect_resolve.ConnectorResolver.capture_context`

Changed
- SubTaskFailed exception to include traceback info
- TaskPartition to require the class. This makes it easier for one model to run another. The simple version of `partition_slice` defaults to the current class.
- AbstractProcessPool.run_subtasks to no longer take the model class as this is now in TaskPartition updated: no longer serialising TaskPartition to json. Instead allowing python's queue to do this in LocalProcessPool.

Changed
- PartitionedModel.partition_initialise to only take key word args, not list of args
- PartitionedModel.partition_plea to be able to return list of TaskPartition objects which allow for richer initialisation of the worker
- (breaking change) - interface from AbstractProcessPool.run_subtasks to take TaskPartition objects instead of tuples.
- single process execution in PartitionedModel is now more like multiprocess execution as the model is constructed and partition is initialised for each sub-task.

Removed
- PartitionedModel.worker_initialise - it doesn't do anything that can't be achieved with PartitionedModel.partition_initialise plus .partition_slice